Fixes #1509

Since #1005 every generated multipart/form-data request sets Content-Type: multipart/form-data; boundary=+++. httpx2 builds the body with that boundary, so a file that contains \r\n--+++ is split there and the server gets a truncated file plus a stray part.

This stops setting Content-Type for multipart/form-data bodies. httpx2 then adds the header with its own random boundary, the same way single-body endpoints worked before 0.29.0. For endpoints with several body types, the multipart branch no longer sets the header either, so #947 stays fixed: httpx2 still writes a boundary because the header isn't set. The other content types still set their header explicitly.

Changes:

• endpoint_module.py.jinja: skip the Content-Type line for multipart bodies.
• Golden records and the integration-test client regenerated (only the boundary=+++ line goes away).
• New functional test test_multipart_bodies.py: generates a single-body and a multi-body endpoint, sends a file containing \r\n--+++ through a MockTransport and parses the multipart body back.
• Patch changeset.

Testing:

• pytest end_to_end_tests/functional_tests/generated_code_execution/test_multipart_bodies.py
  • before the template change: 2 failed ({'file': b'first line'} != {'file': b'first line\r\n--+++\r\nstill the same file\r\n'}), 1 passed (JSON control)
  • after: 3 passed
• Full pdm test equivalent with coverage: 471 passed, 100% coverage, 5 snapshots passed
• ruff format --check ., ruff check ., mypy openapi_python_client fuzz pass

I didn't run the integration job locally because it needs the openapi-test-server container.