feat: cancel common terms in `Nat` subtraction in the `Sym.simp` `grind` normalizer (#15884)

This PR makes the `Sym.simp`-based `grind` normalizer
(`backward.grind.normalizer := false`) cancel the common part of both
sides of a `Nat` subtraction, as the legacy normalizer does with
`Nat.reduceSubDiff`: `n + 1 - 1` becomes `n`, `n + 2 - 1` becomes `n +
1`, and `3 - (n + 1)` becomes `2 - n`. Common monomials are cancelled
too, so `n + k - k` becomes `n` and `x * y + 3 - (x * y + z + 1)`
becomes `2 - z`.

Without this, `[:n].size` unfolds to `n + 1 - 1`. Cutsat knows that this
equals `n`, but the E-graph does not, so goals where the two terms occur
under binders could not be closed by congruence. The new `post` simproc
`Sym.Simp.simpNatSub` calls `Sym.Arith.cancelNatSub?`, which normalizes
both sides, takes the monomial-wise minimum `c` of their polynomials,
and rewrites `(a' + c) - (b' + c)` to `a' - b'` using `eq_normS` and
`Nat.add_sub_add_right`. As with other facts the normalizer proves,
`grind` now rejects theorems such as `Nat.add_sub_cancel` as having no
usable pattern.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
4 files changed