React Compiler: Store aliasing values in an inline set - #37366

97.6% of the value sets tracked per identifier in mutation / aliasing
inference hold exactly one element, but each was a `FxHashSet`, meaning
each was a heap allocation.
Because the inference code retains a full state in each basic block,
these single-element hashsets were a major contributor to peak memory
allocation.

This replaces them with a small inline set inspired by smolvec / tinyvec.
Five values are stored inline, and any more spill to the heap. This was
only needed in **0.02%** of sets in my data corpus.

This also makes iteration order match the TS implementation. TS uses
`Set` and iterates in insertion order; the Fx set iterated in hash
order. This brings the two behaviors in line.

| Benchmark        | Peak allocation            | Allocation count | Wall time |
|------------------|----------------------------|------------------|-----------|
| legacy/image.tsx | 58.47 -> 53.25 MiB (-8.9%) | -31.6%           | -21.1%    |
| next-client      | 58.47 -> 53.11 (-9.2%)     | -25.0%           | -14.0%    |
| devtools         | 26.58 -> 24.52 (-7.8%)     | -15.6%           |  -8.6%    |
| fixtures         | 17.47 -> 15.95 (-8.7%)     |  -7.8%           |  -3.1%    |
| next-examples    |  6.22 ->  5.98 (-3.9%)     |  -5.2%           |  -1.1%    |

Building on the `ValueIdSet`, let's store the spilled values in a
boxed slice instead of a Vec. A Vec has 32 bytes, a Boxed slice
only has 24. This is another 7-9% memory savings on top of the
previous commit.