perf: remove pointer-dereference using owned version of visit_operator

[dishmaker]

This changes Visitor from &mut V to V, in order to reduce pointer dereferences.

I did also optimize skip_const_expr, which showed up in flamegraph of:
wasmtime compile foo.wasm from bytecodealliance/wasmtime#14260

     Running benches/benchmark.rs
parse/tests             time:   [7.4698 ms 7.4766 ms 7.4874 ms]
                        change: [−2.7403% −2.5706% −2.4035%] (p = 0.00 < 0.05)
                        Performance has improved.
Found 3 outliers among 100 measurements (3.00%)
  2 (2.00%) high mild
  1 (1.00%) high severe

validate/tests          time:   [43.371 ms 43.406 ms 43.441 ms]
                        change: [−0.6950% −0.5781% −0.4792%] (p = 0.00 < 0.05)
                        Change within noise threshold.

[alexcrichton]

Thanks! Would you be ok splitting out the skip-const-expr change?

Otherwise for the change here I'm less sure. From an ergonomics/design point of view it seems worse to me to use T instead of &mut T (e.g. impls on &mut T or similar) and the performance gain here seems pretty modest (insofar as validation is the main bottleneck I'd imagine to optimize as opposed to pure parsing). Do you have other changes/improvements that depend on this though?

[dishmaker]

Ok, I will split the changes.

And no, I don't have other improvements.

[dishmaker]

Looks like performance didn't change, so I will close this.

parse/tests             time:   [7.7166 ms 7.7221 ms 7.7306 ms]
                        change: [−0.1755% −0.0743% +0.0518%] (p = 0.22 > 0.05)
                        No change in performance detected.
parse/tests             time:   [7.6719 ms 7.6761 ms 7.6818 ms]
                        change: [−1.6105% −1.4575% −1.3167%] (p = 0.00 < 0.05)
                        Performance has improved.
parse/tests             time:   [7.6793 ms 7.6863 ms 7.6951 ms]
                        change: [−1.1005% −0.9505% −0.8061%] (p = 0.00 < 0.05)
                        Change within noise threshold.
parse/tests             time:   [7.7045 ms 7.7109 ms 7.7184 ms]
                        change: [−0.6210% −0.4954% −0.3695%] (p = 0.00 < 0.05)
                        Change within noise threshold.
parse/tests             time:   [7.7282 ms 7.7378 ms 7.7499 ms]
                        change: [+0.0477% +0.2031% +0.3622%] (p = 0.01 < 0.05)
                        Change within noise threshold.

validate/tests          time:   [44.517 ms 44.549 ms 44.584 ms]
                        change: [+0.9563% +1.0937% +1.2295%] (p = 0.00 < 0.05)
                        Change within noise threshold.
validate/tests          time:   [44.146 ms 44.189 ms 44.239 ms]
                        change: [−0.4287% −0.3231% −0.2004%] (p = 0.00 < 0.05)
                        Change within noise threshold.
validate/tests          time:   [44.129 ms 44.175 ms 44.223 ms]
                        change: [+0.5087% +0.6453% +0.7659%] (p = 0.00 < 0.05)
                        Change within noise threshold.
validate/tests          time:   [44.397 ms 44.432 ms 44.469 ms]
                        change: [+0.6127% +0.7145% +0.8122%] (p = 0.00 < 0.05)
                        Change within noise threshold.
validate/tests          time:   [44.170 ms 44.202 ms 44.236 ms]
                        change: [−0.0238% +0.1051% +0.2291%] (p = 0.11 > 0.05)
                        No change in performance detected.