feat: weighted question scoring with deterministic tie-breaking

[Newstrife]

Problem

determineType uses a hard > 24 threshold. With 8 equally-weighted questions per dimension (score range 8-40, midpoint 24), an exact score of 24 is reachable and silently resolves to the left pole (E/S/F/J) — a coin-flip result for borderline users. The same issue exists in determineQuickType (threshold 6) and determineDimensionPreference (threshold 24).

Solution

• Per-question weights (0.6-1.3) reflecting how strongly each OEJTS item discriminates its dimension (e.g. Q24 "details vs big picture" = 1.3 for SN; Q16 "fits in vs stands out" = 0.6 as a weak proxy). Weights per dimension sum to exactly 8.0, so the score range (8-40), neutral midpoint (24) and percentage mapping are all unchanged — zero downstream breakage.
• Non-uniform weights make an exact tie at the midpoint practically unreachable.
• Deterministic tie-breaking for the remaining exact-tie cases (e.g. all-neutral answers): breakTie examines the dimension's questions from highest to lowest weight and uses the first non-neutral answer, instead of silently defaulting to the left pole.
• determineType / determineQuickType / determineDimensionPreference accept an optional answers param to enable tie-breaking — fully backward compatible; generateResult / generateQuickResult / generateDimensionResult pass it automatically.
• Quick test thresholds and percentages are now derived per dimension from question weights (EI/SN: 7.2, TF: 6.9, JP: 6.75), with percentage clamping to 0-100.
• Weighted scores are rounded to 2 decimal places to keep result URLs clean.
• New exports: questionWeights, getQuestionWeight, dimensionWeightSums.

Tests

175 tests pass, including new coverage for:

• Exact-tie breaking at 24, both directions + all-neutral fallback (full test)
• Exact-tie breaking in the quick test
• Weight integrity: every question has an explicit weight, per-dimension sums = 8.0
• Percentage clamping for out-of-range scores

Notes

• Weights are assigned from face validity of the OEJTS items. If real response data becomes available, they can be re-calibrated via item-total correlations.
• Companion PR for the openmbti.org site: ya-luotao/openmbti.org#20