Performance optimizations and Version#== clarity fix

[andrew]

• Memoize Version#to_s and Interval#empty? since both are immutable after construction
• Optimize Interval#overlaps? with inline bounds check instead of allocating a full intersection Interval
• Single-pass interval filtering in VersionRange constructor (replace compact.reject with select)
• Use match? instead of match in hot paths to avoid MatchData allocation
• Improve cache eviction strategy: drop oldest half instead of clearing everything, bump limits
• Precompile OPERATOR_PREFIX_REGEX constant in Parser#sort_key_for_constraint
• Fix Version#== operator precedence bug (was parsing as self <=> (other == 0) due to == binding tighter than <=>)
• Add benchmark gem for Ruby 4.0.0 compatibility in rake tasks
• Deduplicate Interval empty computation (validate_bounds! and empty? were doing the same comparison)
• Replace fake memory benchmark with real allocation counting via ObjectSpace
• 9 new tests covering equality edge cases, nil interval filtering, overlaps edge cases, and to_s memoization

Speed

Benchmark	main	this branch	change
Native parsing (ranges/sec)	10,331	74,890	7.2x faster
VERS URI parsing (URIs/sec)	11,553	93,664	8.1x faster
Version comparison (ops/sec)	1,200,000	1,565,217	1.3x faster
Containment checks (ops/sec)	121,507	253,165	2.1x faster
to_s conversion (ops/sec)	222,222	1,619,049	7.3x faster
contains? repeated (ops/sec)	111,842	178,010	1.6x faster
empty? repeated (ops/sec)	11,333,467	16,999,871	1.5x faster
npm scheme parsing (ms/parse)	0.0039	0.0026	1.5x faster
maven scheme parsing (ms/parse)	0.0882	0.0231	3.8x faster
Complex npm stress (ops/sec)	45,263	115,088	2.5x faster
Gem pessimistic stress (ops/sec)	150,875	381,534	2.5x faster
Maven bracket stress (ops/sec)	51,403	121,124	2.4x faster
Python complex stress (ops/sec)	25,374	61,728	2.4x faster

Allocations

34 version ranges parsed from test suite data.

Metric	main	this branch	change
Cold parse (total objects)	4,585	1,466	3.1x fewer
Cold parse (per range)	134.9	43.1	3.1x fewer
Cold parse strings	2,756	817	3.4x fewer
Cold parse arrays	1,044	347	3.0x fewer
Cold parse hashes	584	110	5.3x fewer
Cached parse (total objects)	3,478	970	3.6x fewer
Cached parse (per range)	102.3	28.5	3.6x fewer
to_s allocations	725	105	6.9x fewer
contains? allocations	1,473	642	2.3x fewer
empty? allocations	2	2	same
Retained memory (per object)	40 bytes	40 bytes	same