Update lucida.safetensors to v13
Updates the bundled Lucida weights to the current v13 release (source: https://huggingface.co/egeorcun/lucida, commit 9165c2e). The bundled copy is a July-14 snapshot (the v5-era weights); since then the model gained a background-purity training objective and a transparency milestone: overall MAE 0.0257 -> 0.0250 on our 203-image benchmark, transparency 0.0338 (past the commercial reference we test against), camouflage 0.0227, and the community-reported background smears on real photos are measurably reduced (details: https://github.com/egeorcun/lucida/releases/tag/v1.3). Same architecture, same file format and size as the current copy — a drop-in replacement.
Withdrawing this for now — we found a regression on a class of real-world print-design artwork after opening it, and the bundled copy should not move until that is fixed. Will reopen with validated weights.