Inline GC fastpath in ZJIT for newhash instruction - #17717

This extends the ZJIT GC fast path to the newhash instruction for empty
hashes.

This was verified in a stripped down benchmarking environment, using the
following micro-benchmark in ruby-bench:

```
def run(max)
  i = 0
  while i < max
    h = {}
    i += 1
  end
end

require_relative '../harness/loader.rb'

run_benchmark(50) do
  50.times do
    run(500_000)
  end
end
```

This patch resulted in a ~30% speedup.

```
master: ruby 4.1.0dev (2026-07-07T14:04:07Z master c51b159) +ZJIT +PRISM [x86_64-linux]
experiment: ruby 4.1.0dev (2026-07-07T19:27:04Z mvh-zjit-inline-fa.. 60730f1) +ZJIT +PRISM [x86_64-linux]

-----  ------------  ---------------  ------------------  -----------------
bench   master (ms)  experiment (ms)  experiment 1st itr  master/experiment
hash   682.3 ± 0.7%     518.0 ± 1.0%               1.082              1.317
-----  ------------  ---------------  ------------------  -----------------

Legend:
- experiment 1st itr: ratio of master/experiment time for the first benchmarking iteration.
- master/experiment: ratio of master/experiment time. Higher is better for experiment. Above 1 represents a speedup.
```