Add send_rubyfunc_inline benchmark

[k0kubun]

This is a microbenchmark to evaluate ZJIT's PushInlineFrame.

stack map

ruby/ruby#17387 speeds up this benchmark:

before: ruby 4.1.0dev (2026-07-06T20:37:20Z origin/master 733a9df661) +ZJIT +PRISM [x86_64-linux]
after: ruby 4.1.0dev (2026-07-06T21:08:10Z zjit-inline-push f6dfa684fc) +ZJIT +PRISM [x86_64-linux]

--------------------  -----------  -----------  -------------  ------------
bench                 before (ms)   after (ms)  after 1st itr  before/after
send_rubyfunc_inline  36.7 ± 0.8%  34.6 ± 1.9%          1.035         1.059
--------------------  -----------  -----------  -------------  ------------

interpreter vs YJIT vs ZJIT

interp: ruby 4.1.0dev (2026-07-06T21:08:10Z zjit-inline-push f6dfa684fc) +PRISM [x86_64-linux]
yjit: ruby 4.1.0dev (2026-07-06T21:08:10Z zjit-inline-push f6dfa684fc) +YJIT +PRISM [x86_64-linux]
zjit: ruby 4.1.0dev (2026-07-06T21:08:10Z zjit-inline-push f6dfa684fc) +ZJIT +PRISM [x86_64-linux]

--------------------  ------------  -----------  -----------  ------------  ------------  -----------  -----------
bench                  interp (ms)    yjit (ms)    zjit (ms)  yjit 1st itr  zjit 1st itr  interp/yjit  interp/zjit
send_rubyfunc_inline  447.2 ± 0.8%  46.3 ± 0.8%  34.6 ± 1.0%         9.438        12.584        9.655       12.935
--------------------  ------------  -----------  -----------  ------------  ------------  -----------  -----------

[k0kubun]

The ruby-head CI failure seems unrelated. I'll address that in a separate PR.

[tekknolagi]

Is the interp/YJIT/ZJIT before or after the PR you mentioned that speeds us up?

[k0kubun]

after. I generally include the branch name and sha (2026-07-06T21:08:10Z zjit-inline-push f6dfa684fc) when I paste ruby-bench results.