Add entityType discriminator and isolate tables from views

[ruolin59]

Summary

Tables and views share a single (databaseId, objectId) pointer key space, so a
name must resolve to exactly one catalog object. This adds a nullable
entityType discriminator end-to-end — MySQL/H2 → HTS → generated client →
internal HouseTable pointer — and makes every table path aware of it.

Semantics: NULL and any case spelling of TABLE mean table; any case
spelling of VIEW means view; any other non-null value fails closed (neither
table nor view). The column is nullable and deliberately not backfilled, so
existing rows stay valid and ordinary table commits continue to write no
discriminator at all — no change to existing metadata.json content.

Resolves BDP-108403 (consolidating BDP-108404, -108405, -108406, -108409, -108422).

Read paths

Views are filtered in the repository query, including the count query — never by
post-filtering a Page, which would return short pages and inflated totals.

entityType IS NULL OR upper(entityType) = 'TABLE'

IS NULL is required because the backfill is deferred; plain equality would hide
every existing table. upper(...) makes the check independent of provider collation.
The predicate is a shared constant, so each countQuery is the same literal as its
value and the two cannot diverge.

Applied to both /hts query families, listTables and its paginated overload,
listHouseTables, all searchTables overloads, and database enumeration.

Write paths

Two different questions, deliberately separated:

Question	Method	Sees VIEW?
Can this load as a table?	findById / findTableRefById / loadTable	No
Is this name occupied, and by what?	findOccupyingEntityTypeById (new)	Yes

CREATE and rename-destination check occupancy before authorization, allocation,
metadata writes, and pointer saves — otherwise a CREATE at a view's name looks free
and fails only at the HTS publish boundary, after writing an orphaned
metadata.json. HTS 4xx/5xx propagate rather than reading as a free name.

The drop guard cannot live in doRefresh — deleteTable bypasses loadTable so
drops survive corrupted metadata, so the guard sits in findTableRefById and
OpenHouseInternalCatalog.dropTable.

Wrong-type read/drop returns 404, collisions 409, never 400 — the Java/Spark
client treats 400 and 404 identically.

Beyond the ticket's literal scope — please review deliberately

(a) entityType API validation touches services/common. ENTITY_TYPE_REGEX
is added to ValidatorConstants and enforced in
OpenHouseUserTableHtsApiValidator. The ticket did not ask for validation, but a
new API field with an enumerated domain should not accept arbitrary strings, and
the constant belongs beside its sibling definitions. Append-only; no existing
constant or validator branch changed.

(b) Catalog renameTable now rejects any occupied destination. Previously
the catalog performed no destination check at all and relied on the HTS primary
key. It now preflights occupancy, so an occupied destination is an explicit
collision before the transaction opens rather than a constraint violation after
a metadata write.

Deferred: HouseTableMapper.toHouseTable(ViewMetadata, FileIO)

Named in the ticket's scope, but moved to the view-commit ticket.
org.apache.iceberg.view.ViewMetadata does not exist in iceberg-core-1.2.0.20
(verified: 0 classes, vs 1 in 1.5.2.17), and
tables-test-fixtures-iceberg-1.2 loads internalcatalog classes against
Iceberg 1.2. A 1.5-only type on a Spring-managed bean therefore fails during
lookup-method introspection with NoClassDefFoundError — 7 fixture failures —
even though nothing calls it. The overload has zero callers in this change, and
the 1.2/1.5 bean boundary is an architectural decision better made where the
view path is actually built.

Incidental fix: stripOhNamespace null-safety

MapStruct adopts any String -> String method on a mapper as an implicit conversion
and applies it to every String property. stripOhNamespace assumed a non-null key,
which held only because HouseTable had no nullable String field. entityType is the
first, so the mapper NPE'd. Now null-safe — this was a latent pre-existing bug; a null
tableVersion would have triggered it identically.

Test coverage

Written test-first: the full suite landed red and went through review rounds
before any production code, specifically to confirm each test can fail against
the wrong implementation it targets.

• Pre-pagination filtering pinned on 13 distinct paged surfaces, each
  asserting content and page size and totalElements and
  totalPages — totalPages is what kills a post-filter that rebuilds a
  PageImpl with filtered content but the original total.
• Collision tests treat AlreadyExistsException and HTTP 409 as proving
  nothing (both the OpenHouse and Iceberg variants map to 409, and the DB
  primary-key fallback throws the same type). They lean on exact message
  identity, byte-identical pointer state, an unchanged *.metadata.json file
  set, and verify(authorizationHandler, never()).checkAccessDecision(...).
• Case/garbage matrix (NULL/TABLE/table/TaBlE/VIEW/view/ViEw/
  UNKNOWN/"") at both the SQL layer and the Java guard layer. The SQL tests
  are explicitly scoped: they prove the upper(...) normalization, not provider
  collation.
• Backward compatibility: legacy NULL rows visible and fully operable
  throughout; normalTableCommitDoesNotStampEntityType asserts ordinary commits
  add no openhouse.entityType property.

Results — all green

Module	Tests	Failures
services:housetables	153	0
iceberg:openhouse:internalcatalog	124	0
services:tables	530	0
tables-test-fixtures_2.12 (Iceberg 1.2)	8	0
tables-test-fixtures-iceberg-1.5_2.12	compile	✅
openhouse-spark-3.5-itest BranchJavaTest		✅
spotlessCheck + checkstyle		✅

Both published fixture variants were compiled and the 1.2 fixture's local-server
tests run, because HouseTableRepository is a shared interface whose predicates
are inherited by Spring Data proxies in published src/main fixture code.

Rollout

1. schema.sql uses CREATE TABLE IF NOT EXISTS, so it does not migrate an
   existing table. Production needs a separately owned
   ALTER TABLE user_table_row ADD COLUMN entity_type VARCHAR(128) DEFAULT NULL
   applied before any code that writes VIEW. No data backfill is required.
2. Deploy HTS before the tables service. HouseTableRepositoryImpl
   intentionally trusts HTS page content and totals rather than filtering
   locally, so a tables service running against an older, unfiltered HTS could
   leak a view into a listing.
3. A MySQL staging smoke test is still required. H2 cannot certify production
   collation behavior. Insert NULL/TABLE/table/TaBlE/VIEW/view/
   ViEw/UNKNOWN rows and exercise both deployed services before views become
   writable.

Follow-ups tracked elsewhere

• restoreTable is the one write to the shared key space without an occupancy
  preflight, and soft_deleted_user_table_row has no entity_type column. Not
  reachable today (views cannot be created yet), but drop table → create view →
  restore table would clobber the view once views are writable. Tracked on the
  view-commit ticket.