bench(lab): read where a repository stands, and answer with it

[luuuc]

Stacked on #298. Review that one first; this PR's diff is only its own changes.

Problem

A repository's position lives in the operator's head. sense-lab status answers where a campaign stands; nothing answers the question a crank has to answer before it does anything — what is the next phase for this repository, and how many authoring cycles has it already spent.

The old driver answered it with a JSON map of repository to phase, edited by whoever was debugging. That is a decision anyone can change by hand, which made it the one number in the system with no evidence behind it.

Summary

Position is a function of two inputs, and which input a fact comes from is the whole design.

The run tree holds which artifacts exist and the authoring cycle — already a directory name under <campaign>/<repo>/<cycle>/. Derived every time, never stored: a counter beside the fact it counts is a counter that drifts from it.

The attempt records hold the verdict, the table that rejected it and the anchor. Recorded once and never re-derived, because no amount of reading a tree recovers a judgment. One file per attempt, written with O_EXCL — append-only as a convention survives until a bad night, and append-only as an open flag survives the bad night.

Out of the two comes a standing, and the standing is an exit code, because this command belongs in a loop.

Code	Means
0	there is a phase to run
2	-show: a position was printed and nothing was done
3	finished: the repository reached the board
4	parked at the authoring ceiling
5	waiting on a human: a PAY nothing will spend on its own
6	refused: the last verdict is not one its phase can emit
7	refused: the verdict is usable and the artifact it owed is not there

Four and five are the two most worth telling apart: one is the method failing on this repository, the other is the method succeeding and waiting on a wallet. Six and seven are diagnosed differently too — an agent that misbehaved, against an agent that died.

-show stops at the announcement, which is the point before an admission writes anything, so it is a preview of what is about to happen rather than a second reading of the same inputs.

Changes

• lab/internal/position: the attempt record and the reader, with a committed fixture campaign carrying one repository of each shape — mid-cycle with two rejections carried, parked at the ceiling, waiting at a PAY, and one that banked a cycle and re-entered
• lab/internal/cli/repocmd.go: the standing, the codes, and -show
• lab/README.md: the code table, and what is derived against what is recorded

sense-lab status is untouched: it stays campaign-wide, and there is no board across repositories here.

Test plan

• make ci green: coverage floor with no new exception, zero complexity suppressions, lint clean
• new files hold 96.4% and 98.2% line coverage, every function covered
• ten mutations that the first draft of the tests let through are now killed, each verified by making the mutation and watching a named test fail: dropping the graph-order sort, checking the phase directory instead of the artifact it owed, treating any banked cycle as a finished repository, keeping only the last rejection instead of all of them, treating a file named like a cycle as a cycle, never assigning the reached phase or the last verdict, letting the count to the ceiling go negative, and dropping the anchor from the record
• every exit code in the table is asserted, and the loop is asserted to stop on each of the four states that must stop it

[luuuc]

reopening so CI runs against main