[naga] Remove non-essential override references via compaction

[andyleiserson]

Adds a mode to compaction that removes unused functions, global variables, and named types and overrides. This mode is used everywhere except the compaction at the end of lowering, where it is important to preserve unused items for type checking and other validation of the module.

Pruning all but the active entry point and then compacting makes process_overrides tolerant of missing values for overrides that are not used by the active entry point.

Fixes #5885

Testing
Adds several cases in the validation test to verify that compaction removes references to non-essential overrides. Also enables the relevant CTS tests, which are now passing.

Squash or Rebase? Squash

Checklist

• If Update naga benchmarks and add a compaction benchmark #7715 is merged before this PR, the call to compact within the benchmark needs to updated with the 2nd argument in this PR.
• Run cargo fmt.
• Run taplo format.
• Run cargo clippy --tests. If applicable, add:
  • --target wasm32-unknown-unknown
• Run cargo xtask test to run tests.
• This change should have a CHANGELOG.md entry.

[andyleiserson]

There is a question what to do about the compact feature. Currently wgsl-in requires compact, so in many environments it is already required. But when wgsl-in is not set, we need a strategy for controlling the use of compaction in process_overrides:

1. Remove the compact feature and build the associated functionality unconditionally.
2. Make process_overrides conditional upon the compact feature, and arrange for the feature to be set when necessary. (Currently the no-features clippy runs in CI are broken, because wgpu-hal uses process_overrides, but does not enable compact.)
3. Make process_overrides available unconditionally, but when built without the compact feature, specifying all overrides is required.
4. Make process_overrides conditional upon a new feature (overrides?) which would depend on compact.

I have listed in my order of preference, but I don't have all the context about use cases and build size concerns that might justify one of my non-preferred options.

[andyleiserson]

It already came in useful to be able to pass the actual entry point here, and I have the changes to do that, but unless there's a desire to have them in this PR, I'll hold them for a follow-up so that this one doesn't get any bigger.

[andyleiserson]

Some performance data: compacting all of the wgsl snapshot tests, with the reachability changes (#7674), before this change:

compact/shader: compact time:   [298.90 µs 299.67 µs 300.65 µs]
                        thrpt:  [472.03 MiB/s 473.57 MiB/s 474.78 MiB/s]

After this change, with !preserve_unused:

compact/shader: compact time:   [336.43 µs 337.32 µs 338.28 µs]
                        thrpt:  [419.06 MiB/s 420.25 MiB/s 421.37 MiB/s]
                 change:
                        time:   [+12.296% +12.679% +13.089%] (p = 0.00 < 0.05)
                        thrpt:  [-11.574% -11.253% -10.950%]
                        Performance has regressed.

Without the reachability changes, after this change, with !preserve_unused:

compact/shader: compact time:   [194.58 µs 195.04 µs 195.50 µs]
                        thrpt:  [691.40 MiB/s 693.02 MiB/s 694.64 MiB/s]

Performance of other stages (with this change; wgsl-in includes a compaction and is ~1% slower, validation and hlsl-out are unchanged):

front/shader: wgsl-in   time:   [8.8573 ms 8.8659 ms 8.8749 ms]
                        thrpt:  [21.790 MiB/s 21.812 MiB/s 21.833 MiB/s]
validate/shader: validation
                        time:   [719.47 µs 720.31 µs 721.13 µs]
                        thrpt:  [268.17 MiB/s 268.47 MiB/s 268.79 MiB/s]
back/shader: hlsl-out   time:   [1.4870 ms 1.5040 ms 1.5299 ms]
                        thrpt:  [92.661 MiB/s 94.255 MiB/s 95.334 MiB/s]