`warnOnUnknownActionKeys`'s dev-console message states a fact step 3 retired, and points the author at the wrong file

[os-sales]

Found while implementing #5611 (declaring overrideNotice on ActionDef). Filed rather than fixed there: different defect class — that card is a missing declaration, this is shipped prose that outlived the change it describes. Not folded into #5611's PR.

The dev-mode warning at packages/core/src/actions/actionKeys.ts:367-373 ends with:

'check for a typo, or promote the key to an explicit field on `ActionDef` ' +
'(packages/core/src/actions/actionKeys.ts). Warned rather than rejected because `ActionDef` ' +
'still carries `[key: string]: any`, so the compiler cannot see this (objectstack#4075).',

Two separate problems, both in text that reaches an author's console.

1. The stated reason is false

ActionDef does not still carry [key: string]: any — objectstack#4075 step 3 deleted it, and actionKeys.pin.test.ts:107 pins the deletion in the opposite direction:

}).toEqual({ ActionDef: false, ActionContext: true });

So the message tells the author the compiler is structurally unable to see an unknown key, while actionDef-closed-surface.test.ts pins that tsc now rejects exactly that at the construction site. The warning's own module header already explains the real reason it survived step 3 — the two mechanisms cover disjoint populations (tsc sees action literals authored in code, the warning sees actions arriving as data from unparsed sys_metadata rows, objectstack#3903). That is the sentence the message should be carrying; it is a genuinely good reason, and the message currently substitutes a retired one for it.

2. The prescription names the wrong file

It says "promote the key to an explicit field on ActionDef (packages/core/src/actions/actionKeys.ts)". The ActionDef interface is at packages/core/src/actions/ActionRunner.ts:112. actionKeys.ts holds the inventory (ACTION_DEF_KEYS), not the interface.

This one has teeth rather than being a typo: an author who follows the path as given edits only the inventory, which is precisely the half-change actionKeys.pin.test.ts goes red on (it re-derives ACTION_DEF_KEYS from the interface's AST, so the entry becomes stale). #5611 had to carry an explicit "steps 1 and 2 must be in the same commit" instruction to keep a dev off that exact rock; the shipped message steers onto it.

Same-root prose elsewhere

Comments that also describe the pre-step-3 world, worth sweeping in the same pass so the file stops contradicting its own pin:

• actionKeys.pin.test.ts:9 — "ActionDef cannot be enumerated at runtime — its [key: string]: any widens keyof to string | number". The conclusion (the list stays data) is still right; the stated cause is retired.
• actionKeys.pin.test.ts:163 — "targt type-checks today: ActionDef's index signature accepts it." It does not type-check today; the sibling closed-surface test pins the rejection.

Suggested disposition

Rewrite the message's tail to give the surviving reason (data-borne actions no compiler ever sees) and point at ActionRunner.ts for the interface, naming the inventory as the second, same-commit edit. Refresh the two comments. No behaviour change — the classification logic is correct and is not in question here.

Refs #5611, objectstack#4075, objectstack#3903, #4046.