06 [grilling] Discrimination test + easy distractor bank

Part of #11

**Blocked by:** #04, #05 _(native GitHub dependency is canonical; this line is a backup)_

## Question

Define the shape-specific discrimination test (§5) with the **easy** distractor bank — the validation centerpiece: convert “gaze looks like it's tracing the shape” into a decodable, falsifiable claim.

**Decisions:**
- **Easy bank**: random other-trial continuations from EC347. Construction + count per test segment.
- **Test**: DTW-align anticipatory-tracing segments vs (a) the true continuation and (b) the distractor bank; does the true continuation win above chance? Win-rate metric + chance definition + null.
- Note: the **hard** confound-matched bank is out of scope for this effort (later effort).

**Resolution:** bank construction + win-rate/chance definition.

No implementation file or test entry point is named. Start by reading the parent issue #11 and the dependency issues #04 and #05, then specify the EC347 easy-bank construction and per-segment count, the DTW comparison, win-rate and chance definition, and the null; done means those decisions are written down for the validation test.

Written by the indexing model from the issue text.