Fix report generation for sessions with no responses and legacy singletarget sessions

[celefthe]

Running visiomode-analysis session on a 2021 singletarget session in which every trial was a miss failed with IndexError: index -1 is out of bounds for axis 0 with size 0. Three problems were stacked: the reaction-time plots called np.percentile on an empty array; singletarget (the pre-rename spelling of targetonly) fell through to the Go/NoGo and 2AFC code paths, so the report tried to plot hit/false-alarm RTs; and the legacy hit/false_alarm/miss outcome labels were only remapped after flattening, so the per-trial SDT inference never matched them and every trial came out unclassified. This branch also merges main back in so the PR does not revert the v0.2.0 release.

Changes

• plots.plot_rt_median / plot_rt_medians_from_dict return NaN markers for empty RT arrays and leave the y-axis to auto-range when the stimulus duration is unknown (legacy sessions without a spec report it as -1 ms, which produced a [0, -0.001] range).
• session.TARGETONLY_PROTOCOLS / session.is_targetonly() centralise the targetonly and singletarget alias; used in _flatten_trials, generate_report, and the report template (via a new is_targetonly template variable). Output filenames keep the original protocol token.
• Legacy outcome labels are normalised per trial inside _flatten_trials (_normalise_legacy_outcome) before stimulus reconstruction and SDT inference; the post-flatten DataFrame remap is kept as a no-op safety net.
• Tests: new write_legacy_singletarget_json conftest factory; regression tests for the empty-RT plots, unknown stimulus duration, the singletarget alias, legacy label normalisation, summary(), generate_report(), and the CLI. 105 tests pass, mypy is clean.
• CHANGELOG: three entries under ## [Unreleased] / ### Fixed.