Improve test pyramid: strengthen unit tests alongside integration tests

[dgenio]

Description

Summary

The current test suite for the Python SDK appears to lean heavily on integration tests:

• many tests spin up full servers/clients,
• numerous tests live under tests/issues/ as end-to-end regressions.

These are valuable, but they make it harder to:

• get fast feedback from focused unit tests,
• refactor internal components without breaking many tests at once,
• pinpoint which component is responsible when a regression occurs.

Proposal

1. Identify core units suitable for isolated testing

   Examples might include:

   • schema generation and func_metadata() helpers,
   • message parsing/serialization,
   • specific error mapping logic,
   • pure utility modules.

2. Add targeted unit tests

   • For each such module, add tests that do not require spinning up a full session or transport.
   • Use mocks/fakes at the boundaries to keep tests small and fast.

3. Keep integration tests, but make them more focused

   • Retain high-value end-to-end tests, especially for bug regressions and cross-component behavior.
   • When possible, scope regression tests to the minimal surface needed to reproduce the bug.

Why this matters

• Developer velocity: Quicker feedback when changing internals.
• Refactorability: Internal improvements (e.g., transport abstraction, session state machine) become safer.
• Signal quality: When tests fail, it’s easier to see which component is responsible.

Acceptance criteria

• Key modules (e.g., schema generation, error mapping) have unit tests that do not depend on full server/client setup.
• Integration tests still cover end-to-end behavior and known regressions.
• New contributions are encouraged to add unit tests where appropriate, not only integration tests.

References

No response