Feat/adaptive stability thresholds

[viktorbeck98]

Task

Description

Segment thresholds and the segment count become configurable instead of hardcoded.

ClassificationMethods gains a segment_thresholds: list[float] field (default
[1.1, 0.3, 0.1, 0.01] — the values that were hardcoded). The segment count is the
list's length, not a separate setting. StabilityClassifier keeps no thresholds of its
own: segment_threshs, n_segments and min_samples are read-only properties over the
block, so reassigning tracker.classification changes all three and a new verdict can be
taken from one ingest.

classify() gains a segment floor after STATIC/RANDOM: a variable with fewer
observations than segments returns INSUFFICIENT_DATA rather than a verdict scored over
empty segments (an empty segment means 0.0 and passes any positive threshold). This is the
only behavioural change at default settings, and it only applies when a segment method is
enabled — a slope-only block cuts no segments and is ungated.

Old snapshots still load: the block wins, a stale top-level segment_thresholds key is
picked up when the block has none, and pre-block state maps through segmentation /
require_declining / incline_threshold as before.

docs/detectors.md documents the field and the floor.

How Has This Been Tested?

45 new tests covering validation, the derived properties, block-swap re-derivation, the
segment-floor table, and each migration row. Full suite: 829 passed, 32 skipped, with the
7 pre-existing failures unchanged. All prek hooks pass, mypy included.

Checklist

• This Pull-Request goes to the development branch.
• I have successfully run prek locally.
• I have added tests to cover my changes.
• I have linked the issue-id to the task-description.
• I have performed a self-review of my own code.

---

^{Need help on this PR? Tag @codesmith-bot with what you need. Autofix is disabled.}