Reduce size of parse error

[kornelski]

Description
Reduces size of parse error from 112 bytes to 48 bytes. This should help with stack usage during parsing, which is especially important given the huge warning about heavy stack usage.

Testing
There's a test asserting the new size.

It has a noticeable performance improvement:

front/shader: wgsl-in   time:   [5.3159 ms 5.3196 ms 5.3245 ms]
                        thrpt:  [291.07 MiB/s 291.34 MiB/s 291.54 MiB/s]
                 change:
                        time:   [-6.7480% -4.1415% -1.9554%] (p = 0.00 < 0.05)
                        thrpt:  [+1.9944% +4.3204% +7.2363%]
                        Performance has improved.

Checklist

• Run cargo fmt.
• Run cargo clippy. If applicable, add:
• Run cargo xtask test to run tests.

[ErichDonGubler]

I'm open to this sort of change, but I noticed a couple of things that I think need to change with this before merges.

I also want to confirm with @teoxoy and/or @jimblandy that this is a desirable change. I suspect that it might get rejected unless we can justify a quantitative gain here in terms of our own benchmarks (which should be easy?). @kornelski: Does this empirically reduce a memory high mark, or improve run times?

[kornelski]

It reduces stack usage of too_many_unclosed_loops test (#5447) from 1.6MB to under 1MB.

[teoxoy]

How does this compare with boxing the error type itself? Maybe it can be done non-invasively via a newtype.

[ErichDonGubler]

My concerns have been resolved. I will defer to the rest of @gfx-rs/naga for merging.

[kornelski]

@teoxoy I've tried boxing errors. It decreases perf by 2%, and the change is way more invasive — needs changing all return types, all return Err expressions, and closures using ? need type hints added.

There are some functions that match on the errors, so to avoid boxing, unboxing, and boxing again, they need to keep returning the unboxed error.

kornelski@c414134

[teoxoy]

I see, thanks for giving it a shot! I think we can land this.