[jihtsan]

Fixes #242.

HTML responses with a closing </head> beyond the first response chunk currently receive no reload script. Keep searching later chunks, retaining at most six trailing bytes so a case-insensitive closing tag split across chunks is recognized. Stop after the first injection; preserve the existing complete-first-chunk path and non-HTML behavior.

For a streaming HTML response whose closing tag is not yet known, remove Content-Length before headers are sent and let Tornado select framing. This avoids the Content-Length overflow described in the Issue without buffering the entire page. If no closing tag appears, all original bytes are emitted.

Validation:

• New regression cases fail on the unchanged implementation, including a real loopback HTTP response with a >64KiB head.
• Python 3.13.12 / Tornado 6.5.10: python -m pytest -q passes 17 tests and 8 subtests.
• Python 3.9.6 / Tornado 6.2: python -m unittest discover -v passes 17 tests.
• Independent verification passes 500 deterministic chunk partitions with bounded pending bytes, and real StaticFileHandler requests for long HTML, no closing head, HEAD, and Range/206.
• git diff --check passes.

Not run locally: Python 3.7/3.8, Linux, Windows, or the Makefile's legacy nosetests command. The repository currently has a tag publishing workflow but no PR test workflow; these results are local validation.