Pass DIVE camera registrations to 2-cam/3-cam VIAME pipelines

[romleiaj]

Multicam pipelines previously received no transform information from DIVE; the old h5 workflow relied on paths hardcoded inside the pipes. Desktop and web now hand the dataset's camera registration to the pipeline's warp processes, mirroring the stereo calibration flow:

• One standard to_registration.json per non-reference camera is written into the job work dir; warpN (matching the pipeline camera position) receives its file via -s warpN:transformation_file=..., with the pair and direction pinned through the reader's from_camera/to_camera config since a pair may be stored in either orientation.
• Pipeline camera order is the registration reference first (the pipes warp everything onto camera 1's frame), then display order. Which detector a pipe runs on which input is the pipe's documented contract, not something DIVE infers.
• Only pairs registering a camera directly onto the reference are supported; pairs between two non-reference cameras are explicitly unsupported for pipelines and never reach the job. Cameras with points-only (unfitted) pairs get no settings.
• Web sends the registration as a multicam_registration job param built from dataset meta; the worker writes the files.

Requires VIAME's dive transform_2d_io reader and warp_detections process.

[romleiaj]

Paired with VIAME/VIAME#278.

[romleiaj]

Pass DIVE camera registrations to 2-cam/3-cam VIAME pipes

Multicam pipes got no transform info from DIVE; paths were hard-coded in the
pipes. Which dataset camera fed which inputN was guessed from display order,
so the thermal detector could silently run on the wrong camera.

This PR sends the dataset's registration to the pipe's warp processes and makes
the camera order an explicit, confirmed step.

Registration hand-off

• One <camera>_to_<camera1>_registration.json per non-reference camera is
  written into the job dir and passed as -s warpN:transformation_file=….
• Camera 1 of the run order is the warp target.
• Only pairs registered directly onto camera 1 are used. Unfitted pairs are
  skipped.
• Files are format v2. VIAME reads only the matrices; the rest is provenance.
• Web sends the registration as a multicam_registration job param; the worker
  writes the files. (Before this, the worker crashed with KeyError: 'a' on v2
  datasets.)
• Needs VIAME dev/dive-registration-json (dive transform reader,
  warp_detections).

Camera assignment

• Pipes can declare slots with # Camera Order: EO, UV, IR. Tokens EO/IR/UV
  match by role (with aliases like rgb/thermal); anything else matches a camera
  name.
• Datasets get cameraRoles (eo/ir/uv), inferred at import from camera names
  (desktop also checks image file names).
• Running a 2-cam/3-cam pipe opens an Assign cameras dialog: one row per
  slot, prefilled by role or name. Empty or duplicate slots block Run.
  Confirming saves roles back to the dataset (checkbox to opt out).
• The confirmed order goes as pipelineParams.cameraOrder. API calls without
  it use the dataset's stored order, as before.

Registration check before the run

• Discovery records which inputs the pipe warps (registrationWarps).
• The dialog shows per row whether that camera is registered onto camera 1 and
  blocks Run with "Register X → Y in the Camera Registration tab first".
• Desktop and web refuse the run with the same message before making a job.

Cleanup

• Removed the old automatic header-matching path (dialog always supplies the
  order now).
• docs/Pipeline-Import-Export.md documents the header and the dialog.

Manual tests

Needs VIAME from dev/dive-registration-json and a registered multicam
dataset. utility_warp_annotations_2-cam.pipe is a model-free test pipe.

• Assign dialog: run a 2-cam pipe with a # Camera Order: header. Verify
  slots prefill by role, swapping works, empty/duplicate blocks Run, roles are
  saved to the dataset.
• Hand-off: with IR registered onto EO, run. Verify the job dir has
  <IR>_to_<EO>_registration.json ("version": 2) and the command has
  -s warp2:transformation_file=….
• Missing registration: clear the IR→EO registration, run. Verify the
  dialog blocks with "Register IR → EO … first"; the API returns the same.
• Web: same hand-off on web-girder. Verify the worker writes the file and
  the pipe finishes.
• No header: run a pipe with no # Camera Order:. Verify bare
  input1/input2 rows in the dataset's stored order.

Stack

Standalone on main. Uses format v2 from dev/auto-register-pipeline.
Companion VIAME branch: dev/dive-registration-json.

Overall: 21 files, +1537/-10. Commits: 8bb338f through 58e1959.