fix: update spatial extension manifest for duckdb 1.5.4

[evan-zhang11]

Problem

PR #340 bumps the DuckDB crate from 1.10500.0 → 1.10504.0 (core 1.5.0 → 1.5.4), but the spatial-extension-manifest.json was not updated. This causes the CI Check spatial extension version sync step to fail.

Fix

Updated backend/extensions/spatial-extension-manifest.json:

• duckdb_crate_version: 1.10500.0 → 1.10504.0
• duckdb_core_version: 1.5.0 → 1.5.4
• duckdb_version: 1.5.0 → 1.5.4
• All 5 platform artifacts updated with new URLs, SHA256 hashes, and local paths

Verification

• Downloaded all 5 platform extensions from extensions.duckdb.org/v1.5.4/ and verified SHA256 checksums
• check_spatial_extension_version.sh passes locally against the updated Cargo.lock

Merge Strategy

This PR targets dependabot/cargo/cargo-minor-patch-4a9f84f695 (PR #340's branch). Merge this first, then #340 will have the manifest fix included and CI should pass.

[evan-zhang11]

CI Update — Version Sync Fixed, but DuckDB 1.5.4 Has Breaking Behavior Change

✅ The spatial extension version sync check now passes with the updated manifest.

❌ However, the actual tests now reveal a breaking change in DuckDB 1.5.4 itself:

Error

TransactionContext Error: Current transaction is aborted (please ROLLBACK)

Root Cause

At backend/src/db.rs:100-115, several schema migration statements use let _ = conn.execute(...) which silently discards errors:

• ALTER TABLE fonts ADD COLUMN ...
• CREATE UNIQUE INDEX ...
• DROP INDEX ...

In DuckDB 1.5.4, one of these statements likely fails (possibly because the column/index already exists from a previous migration), which aborts the transaction. The next call at line 116 (recover_detached_workspace_members) then fails because the transaction is in an error state.

Test Results

• backend_tests: 87 passed, 14 failed (lib), 158 failed (api)
• All failures cascade from the transaction abort in the DB initialization code

Impact

This affects PR #340 as well — it's not just a manifest issue but a DuckDB 1.5.4 behavioral change.

Recommendation

The let _ = pattern on schema migrations should be replaced with proper error handling (check if the error is "already exists" and ignore only that case). Alternatively, wrap each migration in its own transaction or use BEGIN TRANSACTION ... COMMIT with explicit rollback handling.

This needs code changes before PR #340 can merge.