Transition from interpreter to JIT reallocates frozen strings

[headius]

When we run code with frozen strings in the interpreter and then later run the same code as JIT-compiled, those frozen strings objects will be allocated a second time.

This leads to failures in some tests and specs that expect exactly one construction of the frozen string, such as this one in CRuby:

  3) Failure:
TestHash::TestSubHash#test_NEWHASH_fstring_key [/home/runner/work/jruby/jruby/test/mri/ruby/test_hash.rb:352]:
Expected "ABC" (oid=12312) to be the same as "ABC" (oid=12316).

There's no simple fix for this currently, since it's not possible (or at least not easy) to pass already-allocated objects into the cache fields that jitted code uses.

[headius]

Similar issue with literal ranges: #8019