fix: avoid failover on client-local HTTP errors

[abhinavrastogi2627]

Summary

Fixes #143 by keeping client-local httpx failures out of the node-health retry path. Pool exhaustion, local protocol errors, response decoding errors, and redirect-limit errors now propagate without marking a healthy server node unavailable.

The async source remains the source of truth; the sync client was regenerated with utils/run-unasync.py.

Validation

• pytest tests/api_call_test.py -q (33 passed)
• utils/run-unasync.py --check
• git diff --check

A full pytest -q run requires a local Typesense server and also encountered existing unrelated fixture failures in this environment.

[tharropoulos]

thanks for this! i pushed one commit on top with extra tests: the same check for all four client-local errors (not just PoolTimeout), for both the sync and async clients, raised through httpx via respx rather than by mocking make_request. your fix commit is unchanged.