Add Pyramid game variant

[austin-smith]

Closes #22

What Changed

• New Pyramid variant: 28-card seven-row pyramid, draw-one stock with two recycles (three passes), remove exposed pairs totaling 13 (Kings alone), cover-pairs allowed, win by clearing the pyramid.
• Engine: dedicated pyramid/discard/wasteRecyclesUsed state fields with saved-game back-compat, new Selection/Destination cases for pair removal, variant-aware win detection, all mutation through one pure PyramidGameRules.stateByApplying shared by the session and the move advisor.
• Solver-backed hints: PyramidPlanner runs an exact two-stage search over a collision-free 59-bit packed position — weighted A* with an admissible heuristic and a partner-count prune that can prove deals unwinnable, then a max-clear stage so lost deals still get best-effort hints (silence is reserved for provably futile positions). Cached and ratcheted through HintPlanner like Yukon.
• UI: new triangular board view with exposure-driven interaction, stock/waste reuse from Klondike, inert discard pile, win cascade generalized to launch from the discard.
• Also fixes a latent restore bug: loading a game whose variant differs from the stored setting re-dealt over the restored board after hydration.
• Pyramid is wired into tools/hint-probe (deal, hint follower, random control) per its acceptance recipe; the probe's state fingerprint gained a pyramid/discard section separator to avoid false revisit reports.

Why

Issue #22 requests the Pyramid variant. The sources it links disagree on rules; this implements the common digital rule set (pyramid-only win, three passes, cover-pairs) — choices and rationale are documented in docs/solitaire-rules-pyramid.md.

Validation

• 180 unit tests pass (55 new across geometry, rules, session, persistence, and planner, with probe-verified pinned seeds).
• Hint-probe ledger: following every hint wins 80.2% of 500 seeded deals vs a 15.2% random-control floor, zero stalemate-loops, median 67-move wins. All existing variants reproduce their ledger figures exactly, confirming the shared-core changes are behaviorally inert for them.
• Solver verdict sweep at default budget: 79.5% of deals proved winnable, 0.8% proved unwinnable, 19.8% undecided at budget (still served best-effort lines).
• Verified on iPhone simulator end to end: staged fixture restore, pair/King removal with scoring, tap-select-then-tap-partner, hint highlight, stock draw and recycle limits, undo, and the win cascade.

UI Changes

New Pyramid board (seven overlapping rows with per-card drop targets), Pyramid top row (stock, single-card waste, discard pile), Pyramid entries in Settings, Statistics, and Rules & Scoring, and an App Store screenshot fixture.

[austin-smith]

@codex review

[chatgpt-codex-connector]

Codex Review: Didn't find any major issues. More of your lovely PRs please.

Reviewed commit: ad54077add

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[austin-smith]

@codex review

[austin-smith]

@codex review

[chatgpt-codex-connector]

Codex Review: Didn't find any major issues. 👍

Reviewed commit: 9d2883d953

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".