Port 1.9 GC heuristics to v1.12.1+RAI

[d-netto]

Summary

This PR ports the 1.9 GC heuristics code (also used by 1.10.2+RAI) to 1.12.1+RAI. For context: Julia 1.12 uses a variant of MemBalancer with a few guardrails added by this PR.

Correctness Assessment

I was able to build Julia with this patch. I also ran the GCBenchmarks to assess correctness and didn't observe any correctness bugs.

Performance Assessment

I ran the GCBenchmarks with this PR, and compared its performance with our branch of Julia 1.10 (1.10.2+RAI). It's not an apples-to-apples comparison, but it should let us check if GC performance hasn't regressed significantly between these two versions.

I summarized GC time for each benchmark in the table below. All benchmarks were run with a single mutator and GC thread:

----------------------------------------------------------------------------------------------------
| Benchmark                 | GC Time 1.10.2+RAI (ms)  | GC Time 1.12.1+RAI (ms)                  |
|---------------------------|--------------------------|------------------------------------------|
| append.jl                 | 72.2861                  | 94.7843                                   |
| issue-52937.jl            | 618.7                    | 687.381                                   |
| list.jl                   | 1262.47                  | 888.537                                   |
| many_refs.jl              | 844.966                  | 899.122                                   |
| mergesort_parallel.jl     | 935.442                  | 9068.74                                   |
| mm_divide_and_conquer.jl  | 5.45808                  | 8.29879                                   |
| objarray.jl               | 23284.6                  | 14739.2                                   |
| pidigits.jl               | 382.005                  | 468.098                                   |
| pollard.jl                | 134.255                  | 143.216                                   |
| single_ref.jl             | 322.017                  | 364.897                                   |
| strings.jl                | 2185.21                  | 2723.53                                   |
| tree_immutable.jl         | 624.671                  | 539.895                                   |
| tree_mutable.jl           | 1501.02                  | 1549.64                                   |
| tree.jl                   | 384.247                  | 358.776                                   |
----------------------------------------------------------------------------------------------------

[NHDaly]

Per @d-netto:

The PR is good enough to build Julia and run the GCBenchmarks, but I don’t know if it’s stable or performant enough to pass the RAICode tests.
I believe it’s fine to merge and start testing on RAICode, but certainly not fine to deploy to production :).

[github-actions]

This PR is stale because it has been open 30 days with no activity. Comment or remove stale label, or this PR will be closed in 5 days.