fix(pm): let T1 express that the REMOVED value accepted everything

`check-widening-tells`'s T1 says "the accept set gains a spelling an author
may now write", and it fired on a diff that takes a spelling away: a key
narrowed out of `z.unknown()` — zod's universal acceptor — into
`z.union([…])`. Every value the union admits was already admitted, and
`'stage=won'` and `42` are refused now, so the accept set SHRANK.

The finding is the control set, not the reasoning: the SAME removal declined
when the replacement was spelled `z.array(z.any())` and fired when the
identical narrowing was spelled `z.union([`. One semantic change, two
verdicts, decided by nothing but the added value's first line.

#16943's replacement budget was EARNED and then refused at the SPEND, because
#17618 lets a closed-set-valued key spend only on three facts, the second of
which — both member lists readable — a universal acceptor can never carry: it
has no member list, which is exactly the case where no comparison is needed,
since the removed set is the universe and every replacement is a subset of it
by construction.

So a second way to supply #17618's own requirement is added — positive,
block-local evidence that the set gained nothing, carried by the REMOVED
value's semantics instead of by a list comparison. Nothing about the added
value's spelling is read.

The class is measured, not assumed: on zod 4.4.3 `z.unknown()`, `z.any()` and
both with `.optional()` accept all eight probe values, while
`z.unknown().refine(…)` refuses seven — so a narrowing chain step disqualifies
the line, and `z.custom()` (0 code occurrences here) is not declared.

Sensitivity is unchanged and bracketed by controls that still fire: a removed
acceptor on a different key, a removed value that was never universal, a
`.refine()`d one, the dark control with nothing removed, an inline enum
widened in place, and a genuine new key riding along on an exhausted budget.

Claude-Session: https://claude.ai/code/session_017ef78bLdybu3AffehKkhfk
Co-authored-by: Claude <noreply@anthropic.com>

Author: claude