Add skills to compare assembly code

[bitfaster]

Add a dump-asm skill to make it easy to generate baselines for branches and compare the assembly code. This PR adds 3 skills to enable this:

• fast-bench: runs specified benchmark to generate asm files
• split-asm: generate an asm.md file per benchmark method
• dump-asm: generate asm.md files for benchmarks, and store them as baseline data in a directory based on the branch name

The benchmark jobs have been refactored to enable:

• Running via shift+F5 and benchmark switcher, which runs .net48 and .net9.0
• Running at command line with no runtime specified, runs .net48 and .net9.0
• Running at command line with runtime specified, uses specified runtime (tested .net48, .net9.0, and .net48+.net9)

Example command line:

dotnet run -c Release --project BitFaster.Caching.Benchmarks --framework net9.0 -- --runtimes net48 net9.0 --filter *LruJustGetOrAdd* -j short --warmupCount 3 --iterationCount 5 -d --disasmDepth 5

Previously, it was not possible to run at the command line, and specify the runtime AND warmup/iteration count. Runtime would cause the whole bench suite to run, extending runtime from ~2 minutes to ~30 minutes.

[coveralls]

coverage: 99.106% (+0.03%) from 99.072%
when pulling d6e0061 on users/alexpeck/skills
into 20f02a0 on main.