Compiler: auto-size dispatch chunks to stay under JIT HugeMethodLimit

[zklapow]

The JVM's C2 JIT compiler has a HugeMethodLimit (default 8KB bytecode).

At the previous chunk size of 1024 entries, the generated dispatch methods exceed 8KB and never get full C2 optimization. This only affects large wasm modules (>1024 functions), but many real-world modules like CPython, SQLite, etc. easily hit this.

Instead of a fixed chunk size, this PR derives it from the bytecode budget: highestOneBit(hugeMethodLimit / bytesPerEntry), which produces dispatch methods that stay under the limit and get full C2 compilation. The limit is configurable via -Dchicory.hugeMethodLimit=<bytes> for JVMs with non-default settings for the hugemethod limit.

Benchmark results (2000-function module, JMH, JDK 25):

targetFunc	(1024/chunk)	(128/chunk, auto-sized)	speedup
0	0.033 us/op	0.004 us/op	~8x
999	0.034 us/op	0.004 us/op	~8.5x
1999	0.035 us/op	0.004 us/op	~8.75x

In practice in cpython I found this to give about a 15-20% speedup on hot loops.

[zklapow]

I originally did this by simply having MAX_MACHINE_CALL_METHODS be configureable, but that seemed worse than simply deriving from the limit and I imagine most workloads will have left the default 8kb so this produces more optimal results in the base case.

[zklapow]

Will take a look at CI failures

[andreaTP]

Great idea! Thanks, a couple of minor comments/questions.

[andreaTP]

LGTM, thanks!