Use shared httpx.Client in HTTPXTransport for connection pooling

[adamtheturtle]

Summary

• HTTPXTransport was using httpx.request() (module-level function) for every call, creating a new TCP+TLS connection each time with no connection pooling.
• Now stores a shared httpx.Client instance and calls self._client.request(), matching the pattern already used by AsyncHTTPXTransport.
• Added close(), __enter__, and __exit__ methods to HTTPXTransport for proper resource management.
• Added close() to the Transport protocol and RequestsTransport for parity with AsyncTransport.aclose().
• Added a context manager test for HTTPXTransport.

Test plan

• Existing HTTPXTransport tests pass
• New context manager test passes
• All pre-commit hooks pass

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

[cursor]

Protocol's new close() breaks custom transport implementations

Medium Severity

Adding close() to the @runtime_checkable Transport protocol is a breaking change. Since beartype enforces the protocol at runtime via isinstance checks, any existing user-defined transport that only implements __call__ will now fail validation when passed to VWS, CloudRecoService, or VuMarkService. The close() method needs to exist on every Transport implementation, but external consumers have no way of knowing this without a version bump or deprecation notice.