aiohttp: C HTTP Parser Bypasses max_line_size for Fragmented Lines

An attacker sends an HTTP request whose request target (or a response whose reason phrase) is longer than the configured `max_line_size` (default 8190 bytes), but splits the line across multiple TCP segments so that each individual chunk the parser receives is under the limit. Because the C parser checked only the fragment length, the accumulated line could exceed the limit without raising `LineTooLong`, potentially causing excessive memory consumption and a denial-of-service condition. This affects the default C parser used in pre-built wheels.

The bug is in the C HTTP parser's `cb_on_url` and `cb_on_status` callback functions in `aiohttp/_http_parser.pyx`. These callbacks checked only the length of each individual fragment (`length > pyparser._max_line_size`) rather than the accumulated buffer length (`len(pyparser._buf) + length > pyparser._max_line_size`), allowing an oversized request target or response reason phrase to bypass the limit when split across multiple reads.

The patch changes the two checks in `cb_on_url` and `cb_on_status` from `length > pyparser._max_line_size` to `len(pyparser._buf) + length > pyparser._max_line_size`. This ensures the total accumulated line length is compared against the limit, not just the current fragment. The fix aligns the C parser's behavior with the pure-Python parser, which already performed the accumulated-length check.