chore: extend import timing to authoring stages (thumbnails, backdrops, assembler) and record a baseline

[nigel-hall-codes]

Parent: #259 (techdesign) — see sdlc/#259/design.md (2026-07-04 reanalysis), lever L6. Replaces closed #260 as the profiling gate for #263/#264.

Why

#260 (PR #265) added Stopwatch buckets, but no baseline was ever recorded, and every import stage added after it is untimed — it all lands in the "unaccounted" row. On a templated bake the breakdown is actively misleading: the likely #1 cost (per-building thumbnail/backdrop renders + blocking HTTP uploads) is invisible.

What

1. Extend ImportTimings in SFMapImporterWindow.cs with buckets for:
   • buildings sidecar POST (UploadBuildingsSidecar, fix: building thumbnails never upload — import-sidecar never POSTed to server #384)
   • thumbnail render vs upload (RenderAndUploadBuildingThumbnail, Buildings tab thumbnails show placeholder images instead of real 3D building renders #369)
   • facade backdrop render vs upload (RenderAndUploadFacadeBackdrops, feat(server): capture and serve real per-facade (front/rear/side) angle renders for the Building Customization Canvas backdrop #407)
   • BuildingAssembler.Assemble (per-chunk templated pass)
   • BuildingDecalImporter.Import
   • _names.json/_parked.json copy + ImportAsset
2. Run a representative bake twice — authoring server up and down — and post both breakdowns to Speed up full SF map import pipeline #259.

Done when

#259 has real numbers: a per-operation breakdown over a representative import where "unaccounted" is a small residue, with server-up and server-down variants.