Make the inherent impl overlap check linear-time

[jonas-schievink]

Despite #68911, this code was still showing up in profiles, so I turned it from a O(n²) comparison between all inherent impls to an O(n) algorithm that builds an intermediate hash map to find items with the same names. It also makes the code a bit clearer.

The actual performance gains I've measured are pretty small (1-3%), and it's not impossible that the hash map has a negative perf impact in some situations, so I'll query perf for this.

[rust-highfive]

r? @cramertj

(rust_highfive has picked a reviewer for you, use r? to override)

[jonas-schievink]

@bors try @rust-timer queue

[rust-timer]

Awaiting bors try build completion

[bors]

⌛ Trying commit 11c7e82 with merge 696990e8079f8a668b38304d45e9528acaa5aca4...

[bors]

☀️ Try build successful - checks-azure
Build commit: 696990e8079f8a668b38304d45e9528acaa5aca4 (696990e8079f8a668b38304d45e9528acaa5aca4)

[rust-timer]

Queued 696990e8079f8a668b38304d45e9528acaa5aca4 with parent 71c7e14, future comparison URL.

[rust-timer]

Finished benchmarking try commit 696990e8079f8a668b38304d45e9528acaa5aca4, comparison URL.

[jonas-schievink]

Looks like this causes a slight regression in packed-simd, even. If anyone has an idea for improving this, please let me know. If not I guess I'll close this and focus on other optimizations.

[JohnCSimon]

Ping from triage: @jonas-schievink - can you please post your status or close this PR?