Detect response byte-order marks independently of string encoding

[OskarEichler]

Summary

Compare BOM bytes with getbyte so detection does not depend on the String encoding tag. Preserve the supported UTF-16BE, UTF-16LE and UTF-8 signatures and leave the body bytes untouched.

Reproduction

Net::HTTPOK.new('1.1', '200', 'OK').send(:check_bom, ['feff0061'].pack('H*')) returns nil before this change and Encoding::UTF_16BE afterward. The corresponding fffe and efbbbf signatures reproduce the same issue. Actual response bodies arrive as binary strings.

Verification

• Ruby 4.0.6 through rbenv; existing bundle exec rake test: 201 tests, 914 assertions, zero failures/errors, both baseline and this isolated patch.
• 39 focused checks cover all three signatures under four encoding tags, short/non-BOM input, unchanged bytes and complete response-body encoding.
• Ruby syntax and git diff --check pass. Supplemental Lint scan retains the same 25 pre-existing findings; no new findings by file/cop.
• No test/spec files were added or modified, per this contribution's explicit no-new-tests constraint. Focused reproductions were run externally.

Compatibility and limitations

No API or dependency change. Automatic body encoding now recognizes these existing supported BOMs in binary responses. UTF-32 detection is outside this patch.

Based on master 23e859e92e10e43cad520fcacac3fd43640dba7b. Optional external OpenSSL tests and other Ruby/OS combinations were not run. Existing related issues/PRs were checked; this is a focused correctness change, not a claim of exhaustive coverage.