fix(ui-core): the browser's native image drag is impossible on every asset surface

[JArmandoAnaya]

Closes #428

What changed

The browser's native image drag is now impossible on every asset surface, in both modes.

What the render paths actually are — checked, not assumed: the workspace draws the asset as an <img> inside the stage (there is no canvas-drawn path), and product-wide exactly two <img> elements exist — that canvas image, and AssetThumbnail, which the frame-gallery overlay puts inside the workspace and the batch gallery uses for its tiles.

• The canvas image was already inert — draggable={false}, pointerEvents: "none" and userSelect: "none" since the adapter's first commit — so it could not be dragged at HEAD; it gains the chromium tests that hold that true.
• AssetThumbnail was the open surface: its <img> carried no protection, so a drag on a frame-gallery tile (or a batch-gallery tile) lifted a ghost of the picture. It gains draggable={false} — the attribute alone, because an <img draggable={false}> starts no native drag in any engine this product supports, and a second guard behind it would be an untestable double (the AnnotatorPanel precedent).

The e2e stub now gives assets a thumbnail_hash and serves /thumbnail, so the overlay renders real <img> tiles for the test to attack.

Test plan

Both in chromium, with a capture-phase dragstart counter armed before any gesture:

• "no drag lifts the picture, and the tool's own gesture survives the attempt" — a slow full-width drag across the canvas image with the select tool, then a drag on a frame-gallery thumbnail, both counting zero native drags; then an ordinary bbox draw lands, proving the interaction state came through untouched.
• "the read-only mode is no more draggable than the editor" — the same canvas drag on a completed batch.

Harness verification: putting draggable={true} on the thumbnail (the explicit spelling of HEAD's default) turns the first test red — which both proves the counter genuinely observes native drag machinery and is the red-at-HEAD demonstration, since HEAD shipped exactly that state. Reverted by its exact diff; tree byte-identical to the commit.

Gate stages (staged; exit codes verbatim):

Stage	Exit
uv run pytest -q (green except the ten ffmpeg-environment tests recorded on #429)	0
uv run ruff check . / uv run ruff format --check .	0 / 0
uv run mypy src/visionset / uv run lint-imports	0 / 0
pnpm -r build / pnpm test / pnpm -r lint	0 / 0 / 0
openapi drift / generated client drift / mcp reference drift / version sync	0 / 0 / 0 / 0
browser e2e (CI=1 playwright test, 230 tests)	0
browser cycle (CI=1 playwright test -c playwright.cycle.config.ts)	0