Report comparative info for detector scores

[leondz]

Enable interpretation of scores in a run by calibrating them against a bag of models.

• garak/analyze/perf_stats.py takes a glob of report jsonls and calculates mean, standard deviation, and shapiro-wilk p-values (latter is to assess how well the spread of scores fit a normal distribution) for each probe/detector found
• garak/resources/calibration contains the files from which stats are derived in a comparison. These contents are generated from perf_stats.py
• garak/analyze/report_digest.py and templates updated to calculate a z-score for probe/detector combinations where this is possible, given a default calibration json, to print it in the html output, and to also report what this score means & where it came from

[jmartin-tech]

Testing in progress, can you update the description here to explain what this PR does and offer example run of perf_stats.py? Don't need the actual report files just breadcrumbs to follow when we need to update these resources.

[leondz]

Yup, done!

[jmartin-tech]

Due to #813 the resource files here need to replace ContinueSlursReclaimedSlurs80 with ContinueSlursReclaimedSlursMini.

[leondz]

Outstanding catch