Reject sub-$1 tavern bets and fix stale gambling payout dialogue

[dmccoystephenson]

Summary

• Tavern.changeBet only checked affordability, so a negative bet amount passed Player.canAfford (money >= negative is always true) and set currentBet to a negative number. gamble()'s currentBet > 0 guard then silently treated this as "no bet", directly contradicting the "Current Bet: $-100" prompt the game had just shown. Now rejects any amount below $1 with an explicit message.
• Old Tom's tavern dialogue still said gambling would "double your money", but the dice pay out DICE_WIN_MULTIPLIER (5x). Interpolated the constant so the dialogue can't drift from the payout again.
• Adds real-loop regression tests driving Tavern.gamble() itself (loss path, no-bet-placed path) and Tavern.getDrunk()'s money-loss branch, plus regression tests for both fixes above.

Test plan

• python3 -m compileall -q src
• Full suite green: 380 passed (SDL_VIDEODRIVER=dummy SDL_AUDIODRIVER=dummy python3 -m pytest)
• Confirmed each new/changed test fails without its corresponding fix (stashed src/location/tavern.py, reran tests/location/test_tavern.py, saw the new bet-rejection and dialogue tests fail, then restored and confirmed pass)

Closes #127
Closes #126

[dmccoystephenson]

Self-review rubric:

• Scope: PASS — only src/location/tavern.py and tests/location/test_tavern.py touched, both directly implementing Old Tom's tavern dialogue says gambling will "double your money", but the dice pay out 5x #126/Tavern accepts a negative gambling bet, then contradicts itself with "You didn't bet any money!" #127.
• Tests-new: PASS — bet-rejection (test_changeBet_negative_amount_rejected, test_changeBet_zero_amount_rejected) and dialogue (test_npc_tavern_dialogue_matches_dice_win_multiplier) each have a dedicated test; also added real-loop characterization tests for gamble()'s loss/no-bet paths and getDrunk()'s money-loss branch (these were pre-existing uncommitted WIP found in the tree at cycle start — see note below).
• Tests-fix: PASS — empirically verified: swapped tavern.py for the pre-fix version from main, reran tests/location/test_tavern.py, confirmed the 3 new regression tests (dialogue + both bet-rejection tests) FAIL with the expected assertion errors (-100 == 0, 'You have to bet at least $1' not in '...Current Bet: $0', 'double your money' is contained), then restored the fix and confirmed all 29 tests in the file PASS.
• Sibling structure: PASS — new tests follow the # prepare / # call / # check convention and neighbor naming used throughout the file.
• Docs: PASS — no README/PLANNING references to the gambling payout wording or multiplier exist to drift.
• Issue resolution: PASS — Tavern accepts a negative gambling bet, then contradicts itself with "You didn't bet any money!" #127's changeBet now rejects amounts below $1 before the affordability check; Old Tom's tavern dialogue says gambling will "double your money", but the dice pay out 5x #126's dialogue now interpolates DICE_WIN_MULTIPLIER instead of a hardcoded "double".
• CI: PASS — green (https://github.com/Stephenson-Software/FishE/actions/runs/30247045534).
• Money-format: PASS — new rejection message uses %.2f for money, consistent with sibling messages in the same function.
• Deterministic-tests: PASS — all new tests patch random or avoid it entirely.

Note: at the start of this cycle the working tree had uncommitted, never-committed changes to tests/location/test_tavern.py (3 new tests) left over from an interrupted prior session. Two of them (test_gamble_loss_via_real_loop, test_gamble_no_bet_placed) had a latent bug — a mock side_effect that checked an accumulator list's truthiness as a "first call" flag without ever setting it on that branch, so the mocked showOptions returned the same input forever. Running them caused Tavern.gamble()'s while True loop to spin permanently, and the pytest process was observed consuming >19GB of RAM before being killed. Fixed by replacing the flawed sentinel with an explicit call counter in both tests before including them in this PR. Investigating and fixing this seemed safer than either silently discarding untested WIP or shipping it as-is.