Fix empty encoded body sending spurious chunked terminator

[danielalanbates]

Summary

When data encodes to an empty string (e.g. data={'foo': None} or data=[('a', None)]), prepare_body previously set body = '' rather than body = None. This caused prepare_content_length to skip setting Content-Length: 0 (since super_len('') returns 0, which is falsy), and the adapter then treated the request as chunked, sending a bare 0\r\n\r\n terminator that servers like nginx and Apache interpreted as a malformed second request — resulting in a 400 Bad Request.

Root cause: In models.py, self._encode_params(data) returned '' (empty string) for data with all-None values. The empty string is truthy enough to bypass the body is not None check in prepare_content_length, but super_len('') returns 0 which is falsy, so no Content-Length header was set. Without Content-Length or Transfer-Encoding, the adapter fell through to chunked encoding and sent b'0\r\n\r\n'.

Fix: Convert an empty encoded body to None (body = self._encode_params(data) or None) so that prepare_content_length correctly handles it:

• For POST/PUT: sets Content-Length: 0
• For GET/HEAD: omits Content-Length entirely (per RFC 7230 3.3.2)

This is the minimal fix suggested by @nateprewitt in the original discussion.

Changes

• src/requests/models.py: One-line change on the _encode_params call
• tests/test_lowlevel.py: Parametrized test covering both GET (no Content-Length) and POST (Content-Length: 0) with empty form data, verifying no chunked terminator is sent

Test plan

• New parametrized test test_empty_urlencoded_form_body passes for both GET and POST
• All existing test_lowlevel.py tests pass (14/14)
• Verified manually: Request('POST', url, data={'foo': None}).prepare() now has Content-Length: 0 and body is None
• Verified manually: normal data={'foo': 'bar'} still works correctly

Closes #6122

This PR was created with the assistance of Claude Opus 4.6 by Anthropic. Happy to make any adjustments! Reviewed and submitted by a human.