fix(learner): stop InjectLearnedData from corrupting saved spawns

Two bugs made saved learner spawns collapse to one pin on every login:

1. The spawn zone-key migration was inverted: it ran
   GetAreaIdByUiMapId(1241) -> Eversong parent areaId 3430 and moved
   Sunstrider coords there (wrong map / NE corner), deduping distinct
   coords via InsertIfNewBucket along the way. Now it converts only
   legacy areaId keys FORWARD to the canonical uiMapId and leaves uiMapId
   keys untouched. Same for the [9]/[5] home-zone fields.

2. Sanitize de-duplicated coords with the flat COORD_GRID (2.0) instead
   of the per-zone grid, collapsing Sunstrider's tightly-packed spawns
   (grid 0.5). Now uses GetCoordGridForZone(zoneId).

Both rewrote learned.npcs in place, persisting the damage to
SavedVariables. Regression test in QuestieLearnerMultiSpawn_spec.lua.

CHANGELOG.md

@@ -41,6 +41,7 @@
 
 ### Bug Fixes
 
+- **[Learner - InjectLearnedData Was Corrupting Saved Spawns]** The real cause of learner pins collapsing to one (and the data appearing to "not persist"): `InjectLearnedData` ran on every login/redraw and **rewrote the saved learner spawns**. Its zone-key migration was inverted — it called `GetAreaIdByUiMapId(1241)`, which returns the Eversong **parent areaId 3430**, and *moved* Sunstrider's coordinates there (rendering them on the wrong map) while `InsertIfNewBucket` silently deduped distinct coords away. On top of that, `Sanitize` re-deduplicated coordinates with the flat `COORD_GRID` (2.0) instead of the per-zone grid, collapsing Sunstrider's tightly-packed spawns (its grid is 0.5). Together these turned seven saved Arcane Wraith coordinates into one pin and persisted the damage back to SavedVariables. Fixed: the migration now converts only legacy areaId keys **forward** to the canonical uiMapId (never the reverse) and leaves uiMapId keys untouched; the `[9]`/`[5]` home-zone fields are normalized the same direction; and `Sanitize` now de-dups with the correct per-zone grid. A regression test (`Tests/QuestieLearnerMultiSpawn_spec.lua`) drives `InjectLearnedData` + `GetNPC` and asserts all distinct learner spawns survive.
 - **[Learner - Sunstrider Shows Every Learned Pin Again]** Learner-recorded spawns on Sunstrider Isle were being collapsed to a single pin: a prior change kept clustering enabled there in learner mode, so several distinct learned coordinates (e.g. seven Arcane Wraith spots) consolidated into one icon — breaking the DB-building workflow where every learned spawn must be visible. Sunstrider (uiMapID 1241) now shows every distinct pin by default in **all** data-source modes; consolidation only happens if the player explicitly raises the Dense Pin Clustering Aggressiveness knob. (The data was correct in SavedVariables the whole time — this was purely a display/clustering regression.)
 - **[Tooltip - Collapse Stacked Ascension Quest-Progress Lines]** (#9) Tooltips could pile up a stack of progress lines for the same objective — `0/8`, `1/8`, `2/8`, `3/8` — because the Ascension server appends a new progress line on every objective update instead of replacing the old one. A new always-on pass (`DedupeAscensionProgressLines`) collapses each stack to its single most-progressed line. It only ever hides a line when the same tooltip holds another progress line with the same objective text and denominator, so it never touches other addons' tooltip lines or Questie's own single-line objectives (unit-tested). This is separate from the opt-in "Hide Ascension quest progress lines" option (#16), which removes such lines entirely.
 - **[Map - Completed Quest Pins Stuck On The World Map Until Reload]** (#9) A completed quest/objective's pin could stay on the world map (while the minimap correctly cleared it) until a `/reload`. Cause: a draw-queue race. The world-map and minimap draw queues are processed independently in `ProcessQueue`, so when a frame was unloaded while still queued (e.g. an objective that completes the same tick its pins were queued), the frame's minimap-queue entry could trigger the deferred `Unload` first — removing both icons and clearing the unload flag — and then a later map-queue entry would re-add the world-map icon. Because `Unload` had already nil'd the `questIdFrames`/`_G` reference, that re-added pin was orphaned and survived until `/reload`. `Unload` now drops the frame's pending draw-queue entries (new `QuestieMap:DequeueFrameDrawCalls`), so an unloaded frame can never be re-added to a map — keeping both maps in sync.