Implement unboxed_ary to avoid repeated access to packed data

[byroot]

Ruby objects use a lot of bit packing to save on memory, and also have multiple layouts, it's particularly true for Arrays which can be embedded, heap allocated, or shared.

This saves memory, but can cause a lot of execution overhead when the same piece of information is packed and unpacked many times. As functions often unpack data for themselves, but then call other sub routines that need to do the same unpacking again.

Instead if we primarily work with a "normalized" representation of the object, that we keep in sync with the actual RArray, we can save quite a bit of overhead as showcase in the benchmark.

Of course applying this to more than a couple functions is a lot of work and it's unclear whether it's worth it or not.

Also, whenever code using this yield or call into arbitrary Ruby code, we have to assume the unboxed representation may have fallen out of sync. So it may not work so well in some cases.

$ make -j benchmark ITEM="rb_ary_push" BUILT_RUBY="./miniruby" COMPARE_RUBY="./miniruby-baseline"
/opt/rubies/3.3.4/bin/ruby --disable=gems -rrubygems -I../benchmark/lib ../benchmark/benchmark-driver/exe/benchmark-driver \
	            --executables="compare-ruby::./miniruby-baseline -I.ext/common --disable-gem" \
	            --executables="built-ruby::./miniruby -I../lib -I. -I.ext/common  ../tool/runruby.rb --extout=.ext  -- --disable-gems --disable-gem" \
	            --output=markdown --output-compare -v $(find ../benchmark -maxdepth 1 -name 'rb_ary_push' -o -name '*rb_ary_push*.yml' -o -name '*rb_ary_push*.rb' | sort)
compare-ruby: ruby 3.4.0dev (2024-12-05T19:11:39Z ary-batch-info https://github.com/byroot/ruby/commit/e9407cf40626ed55e3a4642f7d7dc083a0f97225) +PRISM [arm64-darwin23]
built-ruby: ruby 3.4.0dev (2024-12-05T19:11:39Z ary-batch-info 46757e76db) +PRISM [arm64-darwin23]
warming up..

|        |compare-ruby|built-ruby|
|:-------|-----------:|---------:|
|64*2k   |     848.804|    1.151k|
|        |           -|     1.36x|
|128*1k  |      1.651k|    2.217k|
|        |           -|     1.34x|

Profile before the change: https://share.firefox.dev/3ZIgQsw

We can see lots of FL_* calls, several of them repeated.

Whereas after the change: https://share.firefox.dev/3Zq7nVn

Most costly FL_* calls have been eliminated, as most of the frequent checks have been batched in the uary_* methods.

[byroot]

I think I got bit again by make benchmark, the gains aren't that good:

/opt/rubies/3.3.4/bin/ruby --disable=gems -rrubygems -I../benchmark/lib ../benchmark/benchmark-driver/exe/benchmark-driver \
	            --executables="compare-ruby::./miniruby-baseline -I.ext/common --disable-gem" \
	            --executables="built-ruby::./miniruby --disable-gem" \
	            --output=markdown --output-compare -v $(find ../benchmark -maxdepth 1 -name 'rb_ary_push' -o -name '*rb_ary_push*.yml' -o -name '*rb_ary_push*.rb' | sort) 
compare-ruby: ruby 3.4.0dev (2024-12-05T19:11:39Z ary-batch-info e9407cf406) +PRISM [arm64-darwin23]
built-ruby: ruby 3.4.0dev (2024-12-06T08:47:22Z ary-batch-info 3ac5702064) +PRISM [arm64-darwin23]
warming up..
# Iteration per second (i/s)

|        |compare-ruby|built-ruby|
|:-------|-----------:|---------:|
|64*2k   |     845.438|   949.870|
|        |           -|     1.12x|
|128*1k  |      1.697k|    1.847k|
|        |           -|     1.09x|

It's still substantial, but probably not quite enough to justify such a huge refactoring.