DecksMD 1.7.7

Changes

FSRS algorithm fixes

• Aligned the FSRS-6 implementation with the official py-fsrs reference math — all primitive formulas now match bit-exactly across 1,396 test inputs
• Fixed post-lapse stability calculation: failing a card on Again can no longer accidentally increase its stability (the FSRS-6 short-term cap was missing) — heavily-lapsed cards are now scheduled slightly more conservatively, in line with actual recall behavior
• Fixed the mean-reversion anchor in difficulty updates so long-run difficulty trajectories match the spec
• Existing card data is preserved; only future ratings are affected
• Validated against the 10,000-user open-spaced-repetition dataset (443M predictions); calibration matches actual recall within 1 percentage point