[recipe] Add McPAL — PAL's search configuration with MAC's gradient momentum

[matanbt]

What

Adds mcpal to the Recipe Hub. It crosses two recipes already hosted here,
differing from each by exactly one knob:

	vs pal__sitawarin2024	vs mac__zhang2024
momentum	0 → 0.6	same
n_candidates	same (128)	256 → 128

PAL contributes the search budget — 128 candidates from a top-256 gradient
ranking, single-token replacement, 1.1× oversampling. MAC contributes momentum
on the ranking gradient, m ← μ·m + (1−μ)·grad, μ=0.6.

Uses GCGPlusOptimizer, since PALOptimizer has no momentum. PAL's two
remaining parameters are no-ops under a self-proxy — proxy filtering is disabled
whenever proxy_model is model, and its unconditional retokenization is
use_retokenize=True — both noted in the docstring.

No reproduction tag, per the naming convention: this is a crossover of two
methods, not a reproduction of either paper.

Why

Momentum and candidate budget are the two axes these recipes differ on, and the
hub currently exposes them only in their published pairings — PAL without
momentum, MAC with a 256-candidate budget. mcpal fills the remaining corner,
so the two can be varied independently from a hosted starting point.

Checks

• ruff check — passes on the new file
• ty check — passes on the new file
• ruff format --check — the new file is already formatted

tropt/recipe_hub/__init__.py fails ruff format --check both before and
after this change (verified by stashing), so it is pre-existing. Left
untouched rather than mixing unrelated reformatting into this PR — happy to fix
it separately if you'd prefer.

I could not run the full pytest suite here (no pydantic in the sandbox
interpreter), so CI is the authority on that. The change is additive: one new
module, one registry entry, one README row.

Files

• tropt/recipe_hub/McPAL.py — new
• tropt/recipe_hub/__init__.py — import + "mcpal" registry entry
• tropt/recipe_hub/README.md — table row