fix(coach): a bonus set no longer reads as a stall the athlete never had

[kurktchiev]

readSession in api/coach/core/payload.js is a deliberate duplicate of the one in
frontend/src/lib/progression.js. The two runtimes share no build step, the same
trade-off server.js already makes for effectiveRoutineId. The copy missed issue #233:
the frontend grades only the sets the plan asked for, this one graded every set logged.

So a fourth set taken short of the goal on a clean 3x10 is a hit in the app and a miss
in the Coach's payload. stallCount reads only this copy, so it reports a stall that
did not happen, and the review and the next proposal are built on it. Two such sessions
and the Coach starts suggesting a deload.

The same logged.slice(0, Math.max(1, planned)) as the frontend, in the same place.
count deliberately keeps reading every logged set: extra sets are the dimension
bodyweight work grows once reps hit their ceiling (#33), they just do not decide whether
the prescription was met.

The header comment claimed coach-parity.test.js pinned these rules against the
frontend's originals. It pins three of the five: modeOf, isBw, isPerSide. isWarmupSet
and readSession were pinned by nothing, which is how this drifted without anything
noticing. api/test/payload-parity.test.js pins them now, and it has to be its own file:
it imports the frontend's real readSession, and a vitest test cannot import into a
node:test file or back.

The new test fails against the old payload.js.

🤖 Generated with Claude Code

[DuarteSantos8]

Thanks @kurktchiev, and good catch on the server copy of readSession. The parity test is a nice guard.

Released in v1.3.9: https://github.com/DuarteSantos8/openGym/releases/tag/v1.3.9