lending: release v11.0.2 (in-house codegen)

[codat-sdk-generator]

Regenerated by generate-and-pr run 31175452062 (https://github.com/codat-internal/sdk-codegen/actions/runs/31175452062) from codat-internal/sdk-codegen @ b5f2513. Version 11.0.1 -> 11.0.2, stamped in packaging and RELEASES.md. All blocking gates ran green in the linked run. Merging this PR is the release trigger.

[mcollinscodat]

I got Claude to run an analysis on every single of of the changes and see if there were any unexpected or breaking changes:

Claude's analysis
Nothing unexpected: every deletion in this PR matches the POC's accepted divergence set. Context first: this is the first in-house regen of lending into experimental/in-house-codegen, so the diff is the whole Speakeasy v11.0.1 output being swapped for POC output at once, which is why it's 1,348 files.

What I checked, deletion by deletion:

• 19 deleted src files. 18 are Speakeasy runtime plumbing (basesdk.py, sdkconfiguration.py, types/basemodel.py, 15 utils/* modules), replaced by the POC runtime (api_client.py, configuration.py, rest.py, _httpx_transport.py, exceptions.py, api/, new models). The importable public modules survive: codat_lending.types still exports all 8 Speakeasy names (BaseModel, UNSET, Nullable, OptionalNullable, etc.) and codat_lending.utils still exists with BackoffStrategy, RetryConfig, Retries.
• The one deleted model, propertie_itemref.py. Intentional: the compat plan maps PropertieItemRef to an alias of ItemReference, and I confirmed the PR head's models/shared/init.py still exports both PropertieItemRef and PropertieItemRefTypedDict. Not breaking.
• 51 deleted doc pages. The 47 subgroup docs/sdks//README.md pages (accounts, suppliers, codatlending, etc.) are all on the presence gate's known_missing list at the generating commit; the POC deliberately emits one README per top-level group instead. The 4 deleted model pages (propertieitemref.md, zero.md, two validdatatypelinks pages) belong to types that were renamed or consolidated and aren't in the gate's required list. Docs deletions don't break code anyway.
• Top-level init.py. Nothing is dropped; the diff removes 8 lines and adds 431, and every name in the removed lines reappears in the added ones.

Two judgement calls worth a second look, neither strictly breaking:

1. pyproject.toml swaps dependencies from httpx/httpcore to urllib3 + python-dateutil + typing-extensions. Any consumer who relied on httpx being pulled in transitively will lose it. Expected for the new runtime, but it's the one real behavioural dependency change. (requires-python also loosens from 3.10 to 3.8, which is safe.)
2. The PyPI metadata changes: description goes from "Make credit decisions backed by enhanced financials..." to just "Lending", and readme switches from README-PYPI.md to README.md. Cosmetic, but it's what shows on the PyPI page, so worth confirming that's intended.

The known residual: codat_lending.utils shrinks from ~48 exported names to 3, so anyone importing Speakeasy internals like unmarshal_json or parse_datetime from it would break. That's the bar the project set (internals aren't supported surface, and the drop-in comparator that ran green in the linked run enforces the supported surface), so it's accepted rather than unexpected. The ris of this is low though, because No consumer imports the internals. Across the codat-internal org, the only repo touching codat_lending besides sdk-codegen is supporttooling, and its one SDK-using script imports codat_platform's top-level class and models.shared only (the codat-lending entry in its lockfile has no matching import). A GitHub-wide search for codat_lending.utils, codat_lending.basesdk and codat_lending.sdkconfiguration finds nothing outside the SDK repo itself; the public codat-docs snippets use top-level imports only.