Collect timing data per unit for each phase

[samestep]

This PR adds a --dump-timings flag to the compile subcommand (similar to the existing --dump-mem-usage flag), which collects timing data per compilation unit for each compilation phase. For example, on my 2020 M1 MacBook:

$ bazel build -c opt //toolchain
$ bazel-bin/toolchain/install/run_carbon compile --phase=lower --dump-timings examples/sieve.carbon | tail
...
---
filename:        'examples/sieve.carbon'
nanoseconds:
  lex:             30792
  parse:           25458
  check:           226625
  lower:           1136958
  Total:           1419833
...

Most of the changes are pretty straightforward. There were a couple I wasn't sure about though; let me know if I should change:

• new Timings class in its own file, pretty similar to the existing MemUsage class
• added a timings_ field to the CompilationUnit class
• added a timings field to the Check::Unit struct
• renamed CheckParseTree function to CheckParseTreeInner for ease of timing with early return

[jonmeow]

Thank you! This looks good, just a few small style things for Carbon's codebase.

[samestep]

Done! Interesting that the style here for an optional pointer is the opposite of, e.g., the style in rust-analyzer.

All look good?

[jonmeow]

Looks great!