refactor(markers): use match statements for marker dispatch

[henryiii]

These are cases where pattern matching is natural (AST-like work is what it was
designed for).

Assisted-by: ClaudeCode:claude-fable-5

[notatallshaw]

Slower:

| Change   | Before [3c6e2aab]    | After [50601022]    |   Ratio | Benchmark (Parameter)                                                                                  |
|----------|----------------------|---------------------|---------|--------------------------------------------------------------------------------------------------------|
|          | 3.99±0.03ms          | 4.14±0.01ms         |    1.04 | markers.TimeMarkerSuite.time_constructor [runnervmvrwv9/virtualenv-py3.10-PYTHONHASHSEED0]             |
|          | 2.90±0.02ms          | 3.05±0.05ms         |    1.05 | markers.TimeMarkerSuite.time_constructor [runnervmvrwv9/virtualenv-py3.11-PYTHONHASHSEED0]             |
|          | 3.40±0.03ms          | 3.51±0.03ms         |    1.03 | markers.TimeMarkerSuite.time_constructor [runnervmvrwv9/virtualenv-py3.12-PYTHONHASHSEED0]             |
|          | 3.35±0.02ms          | 3.48±0.03ms         |    1.04 | markers.TimeMarkerSuite.time_constructor [runnervmvrwv9/virtualenv-py3.13-PYTHONHASHSEED0]             |
|          | 3.27±0.03ms          | 3.37±0.03ms         |    1.03 | markers.TimeMarkerSuite.time_constructor [runnervmvrwv9/virtualenv-py3.14-PYTHONHASHSEED0]             |
|          | 786±8μs              | 846±4μs             |    1.08 | markers.TimeMarkerSuite.time_evaluate [runnervmvrwv9/virtualenv-py3.10-PYTHONHASHSEED0]                |
| +        | 535±5μs              | 610±5μs             |    1.14 | markers.TimeMarkerSuite.time_evaluate [runnervmvrwv9/virtualenv-py3.11-PYTHONHASHSEED0]                |
| +        | 630±5μs              | 728±9μs             |    1.16 | markers.TimeMarkerSuite.time_evaluate [runnervmvrwv9/virtualenv-py3.12-PYTHONHASHSEED0]                |
| +        | 592±5μs              | 685±10μs            |    1.16 | markers.TimeMarkerSuite.time_evaluate [runnervmvrwv9/virtualenv-py3.13-PYTHONHASHSEED0]                |
| +        | 546±5μs              | 643±7μs             |    1.18 | markers.TimeMarkerSuite.time_evaluate [runnervmvrwv9/virtualenv-py3.14-PYTHONHASHSEED0]                |

I did some of my own benchmarking and locally found significantly worse slowdowns in specific cases:

MATCH_CLASS replaces specialized isinstance calls. _format_marker goes from 98 to 123 ops, _evaluate_markers from 141 to 210. Over a sample of real markers on 3.12: str(Marker) +53%, Marker.evaluate +11%; on 3.14 +17% and +12%. Isolated, _format_marker on the [[...]] unwrap path is 1.7x slower on 3.12 and 2.9x on six levels of nesting.

[henryiii]

Ahh, didn't think about this being on a hot path. I know several match statements are slower than the matching if statements, often because they do more checks: [a, b] is slower because it verifies that this is a Sequence, while the if version usually just burns an arbitrary iterator if it's given one. This looks pretty equivalent and I don't see any of the known slow cases, so I'm slightly sad it's slower, but not very surprised. I do wonder if it's extra checks, and which ones it might be.

Looks like it was useful anyway in finding some ways to make this faster, so not wasted. :)

[notatallshaw]

match is not well optimized in general, it should be better in 3.15, various optimizations have gone in. Markers do become a hot path when doing universal resolution.