ZJIT: Add a phase to remove redundant patch points

[st0012]

Some patch points, like the upcoming NoSingletonClass one, can become unnecessary when we have gone through optimizations and got more information. So there could be another optimization phase to remove unneeded patch points based on the context.

More context: ruby#14680 (comment)

[tekknolagi]

There could, but they incur very little run-time cost (there are no comparisons in the JIT code). I think this should be a much later prioritization than other projects

[tekknolagi]

An ideal take on this PR might be:

1. Make a new AbstractHeap called PatchPoint that any potentially Ruby-running code writes to and all PatchPoints read from
2. Iterate over all blocks
3. For each block, make a HashSet of PatchPoint as you see them
4. If you see one already in the table, remove it from the current block
5. If you see an instruction that writes to the PatchPoint heap, clear the set of PatchPoints

This will eventually get fully subsumed by LVN/GVN but for now it's a nice cleanup

Tentatively assigning Randy