[build] Add performance testing of PRs vs current

[kaby76]

This PR, for #4013, helps to solve a significant problem in detecting regressions in grammar performance. Changes to a grammar that solve a syntax problem often result in a performance degradation that is unknown until well after the PR is merged. To detect this problem, a workflow is added to perform a statistical analysis of the performance for a grammar, comparing the grammar before and after a change. While no errors should be raised if the testing fails, the workflow will give a "heads up" that people can look at to see whether the PR introduces a regression.

While it does not solve the greater problem of finding ambiguity and full-context parser fallbacks, these problems can be inferred via a statistical analysis of the performance.

Changes

• A new workflow is added: "perf.yml".
• The workflow executes _scripts/perf-changed.sh.
• The workflow passes to the Bash script what the PR modifies. For each grammar modified, the script checks the runtime of the grammar before and after the PR.
• The sample size (i.e., the number of times all tests in examples/ are run) is the minimum of 40 or the number of times one can run the test under 20 minutes total (10 m before + 10 m after, for each grammar changed). For example, if the test normally takes 3 minutes, then only 10/3 = 3 iterations can be done. The test driver always outputs a "Total Time", which is stored in files for the PR and before the PR, which is in units of seconds.
• A bar graph showing the mean time and standard deviations for each grammar/before-after PR. The bar graph is output as TTY since artifacts are generally not available. Note, it is important to note that you can have a "statistical significance" while the bar graph shows the mean times roughly the same and error bars overlapping.
• A Welch's test is computed to determine "statistical" significance.
• We also check the ratio of the mean runtimes to determine "practical" significance. A cut off of 5% is assumed.
• If there is both a "statistical" and "practical" significant difference in performance, the workflow outputs a message to indicate there is a performance drop. See https://github.com/antlr/grammars-v4/actions/runs/8466843554/job/23196490098#step:15:359
• This PR alters the sql/plsql grammar so that the statistical test can be performed. The grammar was reformatted according to the coding style rules for the repo.

[kaby76]

All set.

[teverett]

@kaby76 thanks!