detect_signals() raises TypeError on every phased chart #141

What happens

detect_signals() raises on every chart run with phased=True: X or mR, the combined by=[] chart or a stratified one, called directly or after focus().

import processbehavior as pb

study = pb.formulate(pb.make_design(state=1, seed=42), response='y', time='time',
                     factors=['factor 1', 'factor 2'])
study.execute(chart='X', by=[], phased=True).detect_signals()
# TypeError: unsupported operand type(s) for -: 'float' and 'NoneType'

r = study.execute(chart='X', by=['factor 1'], phased=True)
r.focus(r.strata[0]).detect_signals()          # same TypeError

Why

• A phased chart gives each phase its own centre line, so its statistics are {'N': None, 'center': None, 'lpl': None, 'upl': None, 'limits_vary': True} (the contract since 5c9929c). The per-phase centre and limits are on the rows.
• SignalDetector.detect sends every limits_vary chart to _calculate_per_row_zones (processbehavior/signals/detector.py, ~337–341), which computes data['upl'] - stats['center']. That was written for Xbar/S with unequal subgroup sizes, where the limits move but the centre is one number.
• Phased charts also set metadata['run_rules_applicable'] = False (70503be: "zones assume fixed center"). Only the renderer reads it; result_signals._detect_for_chart does not, so the detector tries to build zones for rules 2–8 at all.
• No test runs signal detection on a phased chart.

Impact

• The chart itself is right: the per-phase limits are correct, and points beyond them are flagged (beyond_limits, used by the plot, chart_table and get_signals()).
• detect_signals() returns nothing. A caller that catches the error, as the processbehavior.com app does, can only report that signal detection failed.

Proposed change

1. result_signals._detect_for_chart passes metadata['run_rules_applicable'] to SignalDetector.detect.
2. When it is False and the rules are the chart-type default, the default set is Rule 1 only. Rules 2–8 are not applicable rather than skipped, so the result is not marked partial on their account. This is how Xbar/S are already Rule-1-only.
3. An explicit rule list runs as given (the existing "explicit rules, no filtering" contract in SignalConfig.get_rules_for_chart). The per-row zones, the run rule and the violation records use each row's own centre when the scalar centre is None.

Related, not fixed here

Calling detect_signals() at the top level of any stratified result raises ValidationError: Missing required statistics: ['upl', 'lpl', 'center'], followed by the list of strata. result.focus(stratum).detect_signals() works. The message could say so.

Context

Found while scoping phased limits for Xbar/S.