feat: add --stats flag for profiling

[no-yan]

Resolves #614

Some changes from the initial issue:

1. Removed CPU time of typecheck, because we cannot get it without patching tsgo internals.
2. removed wall time of lint, because rules run interleaved across files, and showing elapsed time is meaningless.
3. Add "Traversal+overhead" in lint phase to detect non-rule overhead.

Implement note:

• Measure worker CPU time to calc non-rule overhead.
• Move nil checks inside metric functions to improve readability. Doing nil checks at the caller side made the code verbose. However, ruleListener keeps the check at the call site to minimize function calls, as it is a hot path.

Output

Typecheck:
    Program                                                  Wall Time     Files
    tsconfig.json                                               0.061s        16
    inferred program                                            0.012s         3
    ────────────────────────────────────────────────────────────────────────────
    Total                                                       0.072s

Lint:
    Rule                                                      CPU Time
    await-thenable                                              0.087s
    unbound-method                                              0.062s
    no-floating-promises                                        0.041s
    no-base-to-string                                           0.011s
    no-duplicate-type-constituents                              0.001s
    ... 6 more rules (0.000s)
    ──────────────────────────────────────────────────────────────────
    Traversal+overhead                                          0.002s
    Total                                                       0.205s

Summary:
    Category                                                 Wall time
    typecheck                                                   0.072s
    lint                                                        0.071s