Run benchmarks on GitHub actions.

[bicknellr]

This PR updates this repo to run the benchmarks using GitHub actions.

I originally tried the same approach as https://github.com/lit/lit/pull/1967 but found that the import specifiers were being mapped to paths that didn't exist, which caused the benchmarks to time out. I'm not entirely sure why this didn't work, but I suspect it was due to the way that tachometer expects the repo it will benchmark is set up. Particularly, tachometer seems to expect that the package it runs within stays the same and that different versions of the dependencies are what should be effectively benchmarked.

I've updated this repo's benchmark workflow to check out the benchmark package separately so that it can be the root package with lit-element as its dependencies.

I've merged the LitElement benchmarks from https://github.com/Polymer/benchmarks into this repo under benchmarks and made that folder a private NPM package, which seems to be tachometer's preferred setup.

[github-actions[bot]]

📊 Tachometer Benchmark Results

Summary

callback

• shack: unsure 🔍 -3% - +1% (-0.71ms - +0.20ms)
  ^{this-change vs master}

Results

shack

• Browser: chrome-headless 91.0.4472.164
• Sample size: 350
• Built by: Benchmarks #34
• Commit: 655e9e2

Version	Avg time	vs this-change	vs master	vs previous-2.x-release
this-change	23.03ms - 23.55ms	-	unsure 🔍 -3% - +1% -0.71ms - +0.20ms	unsure 🔍 -2% - +2% -0.55ms - +0.36ms
master	23.17ms - 23.91ms	unsure 🔍 -1% - +3% -0.20ms - +0.71ms	-	unsure 🔍 -2% - +3% -0.37ms - +0.68ms
previous-2.x-release	23.01ms - 23.76ms	unsure 🔍 -2% - +2% -0.36ms - +0.55ms	unsure 🔍 -3% - +2% -0.68ms - +0.37ms	-

_{tachometer-reporter-action v2 for Benchmarks}

[bicknellr]

I'm going to merge this into gh-actions and continue working from there.