Detect HTML in partial reads

[oiahoon]

Summary

• detect HTML in leading-byte reads when a UTF-8 BOM, XML declaration, or HTML comments precede the doctype or html element
• keep the trailing </html> signal and require a real tag-name boundary
• cover the 4 KiB partial-read shape used by Active Storage, including a custom-element negative boundary

This removes the dependency on seeing the end of a document when only its first bytes are available.

Fixes #152.

Validation

• bundle exec ruby -Itest test/magic_test.rb -n '/partial|custom/' (2 tests, 13 assertions)
• bundle exec rake (454 tests, 934 assertions)
• ruby -c lib/marcel/mime_type/definitions.rb
• ruby -c test/magic_test.rb
• gem build marcel.gemspec
• git diff --check