Memoize magic comment regex

[vinistock]

Motivation

I was doing some profiling of the indexing process and discovered we were spending about 7-10 seconds inside the magic_comment_regex method.

I don't know why it's so expensive to instantiate that regex, but there's also no point in creating it every time since it's always the same one.

With memoization this is 40x faster than the previous implementation.

Implementation

Just memoized the regex.

[github-actions]

Benchmark results in seconds (slowest at top)

          textDocument/completion average: 0.231211 std_dev: 0.003845
          textDocument/diagnostic average: 0.051916 std_dev: 0.012414
          textDocument/definition average: 0.005973 std_dev: 0.00332
      textDocument/selectionRange average: 0.005252 std_dev: 0.000803
   textDocument/documentHighlight average: 0.003328 std_dev: 0.00036
 textDocument/semanticTokens/full average: 0.003268 std_dev: 0.00043
      textDocument/documentSymbol average: 0.003161 std_dev: 0.000351
        textDocument/foldingRange average: 0.003122 std_dev: 0.00025
            textDocument/codeLens average: 0.003105 std_dev: 0.000238
        textDocument/documentLink average: 0.0031 std_dev: 0.000342
textDocument/semanticTokens/range average: 0.002112 std_dev: 0.000223
               codeAction/resolve average: 0.001797 std_dev: 0.000154
               textDocument/hover average: 0.001761 std_dev: 0.000159
           textDocument/inlayHint average: 0.001727 std_dev: 0.000184
          textDocument/formatting average: 0.001098 std_dev: 0.000349
    textDocument/onTypeFormatting average: 0.001068 std_dev: 0.000134
          textDocument/codeAction average: 0.001056 std_dev: 0.000104


================================================================================
Comparison with main branch:

 textDocument/semanticTokens/full unchanged
textDocument/semanticTokens/range unchanged
      textDocument/documentSymbol unchanged
        textDocument/foldingRange unchanged
          textDocument/formatting unchanged
          textDocument/diagnostic unchanged
        textDocument/documentLink unchanged
           textDocument/inlayHint unchanged
      textDocument/selectionRange unchanged
   textDocument/documentHighlight unchanged
               textDocument/hover unchanged
          textDocument/codeAction unchanged
    textDocument/onTypeFormatting unchanged
               codeAction/resolve unchanged
          textDocument/completion faster by 2.392 %
            textDocument/codeLens unchanged
          textDocument/definition unchanged


================================================================================
Missing benchmarks:

RubyLsp::Requests::ShowSyntaxTree