chore: deprecate wrap_in_fastapi via PEP 702 @deprecated

[aballman]

Summary

• Mark wrap_in_fastapi (and the etl-uvicorn CLI it backs) as deprecated using typing_extensions.deprecated (PEP 702).
• Add a deprecation banner to README.md pointing users at building a FastAPI app directly.
• Add a CHANGELOG.md entry and bump version 0.0.42 → 0.0.43.

Why

The wrapper obfuscates the mental model of what a plugin actually is. By auto-generating routes from a Python function's signature, plugin authors never have to engage with the HTTP contract the platform actually depends on — which routes exist, what they return, when each one is called. Authors end up reasoning about "my run() function" instead of "my plugin's API surface," and that gap shows up the moment the contract evolves and the helper doesn't keep up.

Encoding the deprecation in code rather than tribal knowledge means:

• IDEs and type-checkers (pyright/mypy) show strikethrough at every call site and import, nudging plugin authors at edit time.
• ruff recognizes the decorator via DeprecationWarning.
• Test suites that don't suppress DeprecationWarning (pytest default) emit one warning per call.

Noise expectations

• Production: effectively zero — Python's default warnings filter suppresses DeprecationWarning from non-__main__ modules.
• CI / pytest: one warning per wrap_in_fastapi call. Verified locally: 68 tests pass, 7 distinct call sites emit the warning.
• Editor / type-checker: persistent strikethrough — this is the channel that does the actual nudging.

Test plan

• make check-ruff passes
• PYTHONPATH=. pytest test/ — 68 passed
• Verified DeprecationWarning fires from wrap_in_fastapi call sites in tests