ViSR sprint: UI improvements

[hyperrealist]

ViSR UI work for the sprint, collected in one draft. Each change is its own commit with its tests, so it can be reviewed, dropped or split out on its own. Server side: DiamondLightSource/dls-tiled#19.

atlas:main (top) against this PR (bottom), on the same data: a fly scan half-way, a 2×2 scan, and a single point.

New: choose how the scan is drawn

A toggle above the plots (a303d7b, 1738c92), all four in so we can try them and decide which to keep:

• Setpoint grid: the spec's setpoint grid from /binned, as before.
• Readbacks: a step-sized rectangle at each recorded position (Tiled's /render).
• Gaussian: a beam-sized blob per point.
• Footprints: each point's beam footprint, overlaps split between neighbours.

Where nothing was drawn, the empty stripes show through, so gaps the beam left stay visible. No run records its beam size yet, so Gaussian and Footprints use one scan step and the plot says so.

New: progress, region detail, resizable plots

• Scan progress (d4d5164): a bar above the plots shows the points recorded against the scan's total ("120 of 289 points", or just the count without a spec), and the shift applied to put the readbacks on the setpoint grid, when there is one. Both come from what /binned and /render already return.
• Pick a region (bcd6a45): "Pick region", then drag a rectangle on any plot; all three show just that part of the scan, drawn by /render for that viewport at full resolution in the current mode. A chip shows the region and goes back to the whole scan; a new scan does too. The region's colours are scaled over that region alone.
• Resizable plots (a477abd): drag the line between the plots and the controls to set the plots' height; arrow keys move it, and a double click or Escape resets it. The height is remembered in the browser.

New: tomography reads reconstructions from Tiled

• Reconstructions from Tiled (3a1ee52): the tomography view finds reconstructions in Tiled's analysis container (array nodes with metadata visr.kind: tomography-reconstruction), shows the newest and reads its bytes through /api/data, so no new proxy route is needed. A picker in the controls switches between them, and a status line shows while none is loaded. The viewer takes 3D uint8 volumes, the format the reconstruction workflow writes (as in public/test-data). The mocks serve that test volume through the same Tiled endpoints.

Fixes

• Unmeasured cells are grey stripes instead of black, so a scan in progress no longer looks like dark data (37a70d2).
• A single point (or all-equal values) is drawn at full brightness instead of black (37a70d2); with dls-tiled#19, 1×1 and 2×2 scans get their own grid.
• Cells sit between their bin edges instead of half a cell right and down: the plot was given cell centres, which it reads as each pixel's leading edge (f8b6698).
• The poll after a scan ends now happens, so the last image includes the last points; it was skipped before. A 503 (data file behind the catalog) is retried instead of logged as an error (ad0fea7).

Checked, no change needed: polling already gets 304 Not Modified when nothing changed, as the browser revalidates with the server's ETag.

Open questions

• Stable colours: colours change during a scan whenever a new extreme arrives, since the range comes from the data so far. A fixed range (detector or user), percentile clipping, or the previous scan's range would avoid it.
• Beam size: to be read from the run's metadata once plans record it (start.beam_size).
• Which drawing modes to keep, and what to call them.

Testing: pnpm typecheck, pnpm turbo build lint test --filter=@atlas/visr...; checked in the dev server against the live b01-1 Tiled, replaying a real 17×17 fly scan. A test release from this branch goes on the beamline before it leaves draft.

🤖 Generated with Claude Code