Avoid allocating source lines when counting

[OskarEichler]

Summary

Count source lines without allocating an array and one string per line.

ASCII-compatible strings use a byte-count fast path. Non-ASCII-compatible encodings retain each_line semantics, preserving UTF-16/UTF-32 behavior.

Evidence

For 100,000-line UTF-8 source, measured allocations fall from 100,002 to 1. Alternating local samples on Ruby 4.0.6 and 3.2.11 were approximately 12–16 times faster. This is a focused microbenchmark, not an application-throughput claim.

Verification

• Exact line-count parity covers empty, final/non-final newline, CRLF, invalid-byte, binary, UTF-8, UTF-16LE/BE and UTF-32LE/BE inputs on both Rubies.
• Release/current baseline and combined completion candidate: unchanged suite passes 913 tests / 4,109 assertions / 0 failures / 0 errors / 4 skips on Ruby 4.0.6 and 3.2.11.
• Full RuboCop checks 133 files with zero offenses; syntax, package parity and isolated package load pass.

Compatibility

Returned data and source-map line-count semantics are unchanged. No dependency/version change. Existing tests were not edited under the consumer repository’s no-test-edit policy; focused evidence ran externally. Prepared with AI assistance.