Trial-to-trial reliability metrics for cue and lever epochs

[celefthe]

Closes #173.

• metrics.reliability_metrics and ReliabilityOptions: epoch_correlation, response_fraction, variance_quench, signal_fraction and reliability_n, appended to _metrics-session.csv for go/no-go sessions, for all trials, per sdt_type (_sdt-<type>) and per stimulus (_stim-go, _stim-nogo). Existing columns unchanged.
• Cue- and trial-start-aligned epochs end at each trial's response (misses and correct rejections at the median RT); lever-aligned epochs run from the cue to the push, with the baseline before the cue.
• New options on process metrics and process peri-event --with-metrics: --mask-response/--no-mask-response, --min-rt, --response-sd, --time-warp, --cue-baseline. The aligned event is read from which trials column matches event_time.
• Skipped with a warning when the trials columns are missing, the session isn't go/no-go, or windows are reward-aligned.
• Peri-event report session table shows the new columns with a trial-group selector.
• Tests, how-to docs and changelog entry.