Performance: replace the boxed ArrayDeque operand stack with a flat Object[] stack

[bertysentry]

Fixes #536

Changes

• Replaced Deque<Object> operandStack (ArrayDeque) in AVM with a plain Object[] plus an explicit top-of-stack index: push/pop/clear are now single array accesses the JIT can fully inline, with no head-index wraparound math.
• Dropped the NULL_OPERAND sentinel: the AVM owns all stack accesses, so null operands are stored directly.
• popArguments() now bulk-copies its argument slice with System.arraycopy instead of popping one element at a time (the hot-path note from the issue).
• The array starts at 64 slots and doubles on demand; operand depth is bounded by expression nesting, so it effectively never grows.
• Popped/cleared slots are nulled so the stack never retains stale references (same discipline as ArrayDeque).

Also renormalized mvnw.cmd line endings: commit 8788dfe stored the blob with CRLF while .gitattributes declares *.cmd text eol=crlf, which made the file show as permanently modified on Windows checkouts.

No user-visible behavior change, so no behavior-changes.md entry.

Benchmarks

JMH, 2 forks, JDK 25, Windows 11 (baseline = branch HEAD before this change):

Benchmark	Baseline	After	Delta
AwkScriptBenchmark.tightArithmeticLoop	9,358 us/op	8,091 us/op	-13.5%
AwkScriptBenchmark.sumInputValues	9.64 us/op	8.81 us/op	-8.6%
AwkScriptBenchmark.projectMatchingValues	11.17 us/op	11.52 us/op	+3.1% (noise)
AVMExpressionBenchmark.literalAddition	27.83 ns/op	25.29 ns/op	-9.2%
AVMExpressionBenchmark.fieldAddition	62.18 ns/op	57.14 ns/op	-8.1%
AVMExpressionBenchmark.fieldMultiplication	63.89 ns/op	55.00 ns/op	-13.9%
AVMExpressionBenchmark.fieldRegexMatch	72.08 ns/op	64.07 ns/op	-11.1%
AVMExpressionBenchmark.fieldConcatenation	69.98 ns/op	66.75 ns/op	-4.6%
AVMExpressionBenchmark.fourStringConcatenation	149.41 ns/op	139.08 ns/op	-6.9%
AVMExpressionBenchmark.multiStringConcatenation	74.11 ns/op	67.89 ns/op	-8.4%
AVMExpressionBenchmark.stringConstantStringConstantConcatenation	155.88 ns/op	139.99 ns/op	-10.2%
AVMExpressionBenchmark.constantStringConcatenation	26.01 ns/op	24.02 ns/op	-7.7%
AVMExpressionBenchmark.mixedExpression	928.77 ns/op	881.17 ns/op	-5.1%

Verification

• mvn test: 752 tests, 0 failures
• mvn clean verify site: BUILD SUCCESS, 0 checkstyle / pmd / spotbugs findings

🤖 Generated with Claude Code

[bertysentry]

@codex please review

[chatgpt-codex-connector]

Codex Review: Didn't find any major issues. Can't wait for the next one!

Reviewed commit: 4ced5f743a

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".