v9.0.1940

What's Changed

• build(msix): bake the Store submission record in the run, beside the package by @sebgod in #261
• fix(imaging): an interior drizzle hole is not a canvas ring, filled at load, flood 64 columns at a time (#250); 8.1 by @sebgod in #262
• perf(imaging): a plane stays float[,]; the hot loops stop reading it as [y, x] by @sebgod in #263
• docs(release): the 8.1 What's New is written before anyone dispatches it by @sebgod in #264
• test(viewer): the controller tests run alone, and their waits bound a stall rather than the work by @sebgod in #265
• docs(plans): HDR in the viewer, what the button does and what the driver allows, measured on this box by @sebgod in #266
• feat(viewer): boost and the highlight soft clip share one Tone popover, and it names the HDR it cannot do by @sebgod in #267
• feat(ui): a click lands the caret where it was aimed, and a second click selects the word by @sebgod in #269
• refactor(ui): a gap is the container's, a rule is a Box, and a fit is the engine's by @sebgod in #270
• refactor(ui): the arrows walk the list the tree already declares, and the docks are trees by @sebgod in #271
• refactor(ui): the T0b sweep rows nobody had taken, and the two engine gaps they found by @sebgod in #273
• feat(ui): the hosts route input through DIR.Lib's InputRouter (T1) by @sebgod in #275
• refactor(ui): popovers and sliders are declarations now (T2) by @sebgod in #276
• docs(plans): DIR.Lib 10, a control is declared once and the engine behaves by @sebgod in #268
• P2 E3.4: the skirt ratio, and a prior that tightens stars without taking their skirt (pool + straddling augmentation) by @sebgod in #244
• docs(plans): the dropdown's engine half landed in 9.3, and what it learned by @sebgod in #277
• build(deps): follow the 9.3 chain, and bring tools/ under CPM (8.2) by @sebgod in #278
• docs(plans): the 9.3 chain is walked, so the dropdown's tianwen half is unblocked by @sebgod in #279
• refactor(viewer): the chrome on the engine — T3 P0/P1/P2/P4, and the DPI bug it uncovered by @sebgod in #280
• refactor(gui): the nav rail declares its chords instead of re-stating them by @sebgod in #281
• docs(plans): recount D2, because the first count was wrong four ways by @sebgod in #282
• refactor(ui): focus changes go through the owner, not past it by @sebgod in #283
• refactor(planner): a handoff divider arms its own drag, and SliderHit goes by @sebgod in #284
• fix(tui): seeding a site field is not claiming the keyboard for it by @sebgod in #285
• docs(plans): the two cuts left are one piece, and OverlayOwnsPointer is not in it by @sebgod in #286
• refactor(viewer): one toolbar press policy, set by the host that disagrees by @sebgod in #287
• refactor(viewer): route keys through the engine, and fix every toolbar button under a router by @sebgod in #288
• refactor(viewer): both hosts route presses — D2's last cut is clear by @sebgod in #289
• Against DIR.Lib 10.0: the cuts reach only the tests by @sebgod in #290
• Dropdowns apply the DPI scale once, and tianwen-fits runs end to end in CI by @sebgod in #291
• Drizzle normalises per CFA colour, and every written master is in [0, 1] with true labels by @sebgod in #292
• feat(web): the browser sky map draws the Milky Way by @sebgod in #293
• feat: verified Wikipedia articles and object pictures in the object panels by @sebgod in #295
• Add Claude Code GitHub Workflow by @sebgod in #296
• ci: @claude can build and test; LFS pull and unit-test deps as composites by @sebgod in #298
• Use ArrayPoolHelper instead of ArrayPool.Rent/Return + try/finally by @sebgod in #297
• fix(serial): the terminated-read window is a stated 128 bytes again by @sebgod in #299
• Object photos: the Veil's picture, a photo mark on atlas labels, one designation per panel line by @sebgod in #300
• The thumbnail surrogate gives its heap back when a burst ends (#294) by @sebgod in #301
• ci(claude): the checkout persists no credential, so Claude can push again (#302) by @sebgod in #303
• Convert the remaining object-typed lock fields to System.Threading.Lock by @sebgod in #306
• Seeing forecast: the wind aloft on the forecast, and a seeing row in the planner (P0 + P1) by @sebgod in #305
• Sensor active area, and the dataset bake the last two nights rebuilt (9.0) by @sebgod in #308
• Seeing for OpenWeatherMap profiles, mixing height, detached app launcher by @sebgod in #309
• build: take DIR.Lib 10.2 and the rebuilt toolkit; fix three viewer menu bugs it surfaced by @sebgod in #310
• Leftovers: the night calendar, Open-Meteo nulls, the Equipment wheel, Store notes by @sebgod in #313
• fix(gallery): every card renders honestly, and one tool per job by @sebgod in #312
• fix(stacking): every strategy retains a coverage plane, so a staged master's crop is exact by @sebgod in #315
• Night calendar P3: when the pinned targets are up on each night by @sebgod in #314
• feat(dataset): a resumed bake re-does a session whose inputs or recipe moved, and only that one by @sebgod in #316
• Night calendar P4: the TUI and the browser, and the TUI list's wheel and selection by @sebgod in #317
• docs(curation): the Rim Nebula ASI533 session is filed as IDAS-LPS-D3 (#18) by @sebgod in #318
• docs(curation): a night folder holds one target; ten re-filed by OBJECT (#25) by @sebgod in #319
• docs(curation): groups O to Q, reachability by content digest, a third provisional slug (#19) by @sebgod in #320
• docs(curation): a flat ratio is a filter times a white balance, so undo it first (#19) by @sebgod in #321
• docs(curation): never mutate a tree while a digest walk is enumerating it by @sebgod in #323
• docs: a narrowband filter on an OSC sensor is effectively mono, and the bake builds it as colour by @sebgod in #322
• Research #48: an obstruction has no edge to mask at, and the BAD_ corpus holds three defects, not one by @sebgod in #325
• Atlas hover highlight, an only-with-photo filter, and a second source for the OpenNGC audit by @sebgod in #324
• docs(stacking): the obstruction plan's design section still closed on the rule its corpus refutes by @sebgod in #326
• fix(imaging): the edge walk's trim cap also bounded its search, and past 0.075 it switched itself off by @sebgod in #327
• perf(imaging): a per-pixel map is quantised before it is compressed, and the bad pixel map is kept by @sebgod in #328
• feat(imaging): a camera whose model number hides its die now finds its QE curve by @sebgod in #329
• fix(stacking): one frame-admission rule for stack and the bake, and two bugs in its keep floor by @sebgod in #330
• build(deps): take SdlVulkan.Renderer 7.45, which obeys the queue's image transfer granularity by @sebgod in #331
• The vendor device drivers are their own assembly, so a viewer stops shipping 8.2 MB of them by @sebgod in #332

Full Changelog: v8.0.1716...v9.0.1940