refactor(transport): add the seams the httpx swap turns on

[baraline]

Plan 2 of the 0.4.0 rewrite. Removes every httpx incompatibility while still running on requests, so the transport swap becomes a mechanical change rather than a risky one. No behaviour change; the suite is green throughout.

• .reason is now read through response_reason() in _http.py. requests spells it Response.reason; httpx spells it reason_phrase and has no reason at all.

• Session construction is centralised in build_http_session(verify_ssl=) and build_client_resources takes an optional session_factory. Two things fall out: the SSL policy is applied at construction, which httpx requires (it reads verify only in Client.__init__ and silently ignores a later assignment, leaving verification ON when the caller asked for it OFF); and tests can inject a stubbed transport without patching module globals. The factory is a Protocol rather than a Callable alias so the keyword-only verify_ssl is part of the contract. GLPIV1Session now builds its session the same way.

• Both dynamic verb dispatches are gone. _transport.py used getattr(self._session, method) and _v1_session.py used getattr(self._http, method.lower()); both now call session.request(VERB, url, ...), the one call shape requests and httpx agree on. Worth noting the getattr in _transport.py was also hiding an untyped call: replacing it made mypy reject the **kwargs: object passthrough that had previously type-checked only because getattr returns Any.

• New test_method_invocation.py drives all 85 public methods against a stub installed at the session.request seam, so URL building, headers, parameter normalisation, response validation and model parsing all really run. It asserts each method reaches the transport, that the async client mirrors the surface, and that the surface has not shrunk. Three methods are exempted with stated reasons. This is the regression net for plans 3-5, where the failure mode is a method that stops dispatching entirely rather than one that returns a wrong payload -- a per-endpoint suite only catches that where a test happens to exist.

  A companion test pins the seam by stubbing only session.get and asserting it is NOT used, so a regression to per-verb dispatch fails.

Two items from the plan needed no work, verified rather than assumed: every re-exporting __init__.py already declares __all__, and _config.py at 249 lines did not warrant splitting.

562 tests pass (from 473), mypy strict clean, ruff clean, coverage 96.90%.

[codecov-commenter]

⚠️ Please install the to ensure uploads and comments are reliably processed by Codecov.

Codecov Report

❌ Patch coverage is 95.23810% with 1 line in your changes missing coverage. Please review.
✅ Project coverage is 96.89%. Comparing base (4239e1b) to head (fcd5522).

Files with missing lines	Patch %	Lines
glpi_python_client/clients/commons/_http.py	80.00%	1 Missing ⚠️
❗ Your organization needs to install the Codecov GitHub app to enable full functionality.

Additional details and impacted files

@@            Coverage Diff             @@
##             main      #18      +/-   ##
==========================================
+ Coverage   96.69%   96.89%   +0.20%     
==========================================
  Files          79       79              
  Lines        2634     2644      +10     
==========================================
+ Hits         2547     2562      +15     
+ Misses         87       82       -5     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.