Build 26374469

NEW: Text recognition is lighter on every capture. The batch assembly and decoding steps around text recognition now skip redundant work - same math, same results (verified bit-for-bit identical recognition on 3,000+ screenshots), shaving roughly 20 ms off every scan on a 1080p screen. Stacked on top of v1.74's detection speedup, translations now appear about 4-8% sooner end-to-end on the default GPU setting.

NEW: The dark-scene reading helper decides faster. The brightness statistics it gathers on every capture dropped from up to about 10 ms to about 1 ms, with identical decisions on every tested frame.

Under the hood: both changes sit behind safety valves and the full 12,000+ test gate passed before shipping. If anything behaves oddly, please report it on the community hub - recent threads directly shaped this update.