Analysis ignores everything the roach ecology now records

[martymcenroe]

Problem

The Andvari round now records a great deal more than it did, and silphe.analysis reads none of it. session_signature still pulls exactly one thing out of an evasive round — path, correlated against the cursor for pursuit lag — which is all it did when a round was one roach on an open field.

Sitting unused on every evasive record:

• roaches[*].modes — the mode timeline (wander, fleeing, baited, feeding, hidden, tunnelling). Chasing a roach drawn toward a crumb and chasing one bolting from you are different measurements, and right now they are averaged together into a single lag number.
• roaches[*].tunnels — dive and surface timestamps. The window between them is the player losing the target, and the cursor trace just after a surfacing is re-acquisition: finding it again and re-homing. That was the whole point of Tunneling: roach drops into a tunnel and re-emerges elsewhere (re-acquisition) #46 and nothing computes it.
• target_switches — when attention moved between roaches. How fast someone re-homes after their quarry dies or escapes is a genuinely new motor measure and is currently only stored.
• maze — the field layout. Anticipating a corner and reacting in the open are different measurements; without folding this in, they are pooled.
• bait, gecko — where the attractors were, and a robot pursuit trace with no human hand in it.

The data has been captured correctly since each feature landed. Nothing reads it.

Wanted

Extend session_signature (or add alongside it) so the evasive round yields segmented measures rather than one pooled lag: lag by roach mode, re-acquisition time after a tunnel surfacing, target-switch latency, and corridor-versus-open pursuit.

Notes

• The schema is documented in docs/0005-session-schema.md; all of these are additive fields under SCHEMA_VERSION 1.
• Worth doing before there is much recorded data to reprocess, but it does not block the release in Release the roach ecology to PyPI as 0.2.0 #62.