Eliminate lazy initialization overhead from interpreter hot paths (~8% faster)

[lahma]

Summary

• Remove _initialized bool field and branch from JintStatement.Execute() — the interpreter's hottest path — by moving all Initialize() work into constructors
• Apply the same pattern to 12 expression types (JintBinaryExpression, JintCallExpression, JintMemberExpression, etc.)
• Remove EvaluationContext dependency from ExpressionCache.Initialize() so call/new/array/object expression caches can also be built eagerly
• Move JintStatementList and JintSwitchBlock array construction to constructors
• Net result: 36 files changed, 212 insertions, 468 deletions (simpler code that's also faster)

Rationale

Every JintStatement.Execute() call checked if (!_initialized) on every invocation — not just the first. In tight loops this branch accumulates measurably. Since none of the Initialize() methods actually used the EvaluationContext parameter (they only called JintExpression.Build(), JintStatement.Build(), and ProbablyBlockStatement()), moving the work to constructors is safe and eliminates the per-execution branch.

Benchmark results (SunSpider, 26 tests, lower is better)

Phase	Total	vs Baseline
Baseline (main)	2,693 ms	—
+ Statement _initialized elimination	2,516 ms	-6.6%
+ Expression _initialized elimination	2,490 ms	-7.6%
+ StatementList/SwitchBlock init	2,474 ms	-8.2%

Biggest individual improvements: bitops-nsieve-bits -10%, date-format-xparb -10%, math-cordic -8.9%, crypto-aes -7.8%, crypto-sha1 -7.1%

Test plan

• Jint.Tests: 2,764 passed, 0 failed
• Test262: 95,798 passed, 0 failed, no regressions
• SunSpider benchmark: ~8% improvement
• Dromaeo benchmark: no regressions
• Stopwatch benchmark: no regressions

🤖 Generated with Claude Code