Allow users to define canonical experiment channel names and record them in the run manifest #170

Problem

Preprocessing currently records per-file channel names only when image metadata provides them. They are observational metadata in each FileEntry; unnamed channels remain unnamed.

The notebook should let the user explicitly name the experiment's channels when metadata are missing or incomplete.

This is also needed before we can sensibly verify/channel-map heterogeneous acquisitions.

Design rule

Keep source metadata separate from user-stated experiment identity.

Do not overwrite FileEntry.channel_names: those fields should continue to mean “what this source file's metadata said”.

Add an experiment-level canonical channel schema, for example conceptually:

index   name        source
0       DAPI        stated
1       ACTIN       stated
2       PAXILLIN    header

The exact storage can be decided during implementation, but it should be reproducible and editable. A small input file such as channels.csv / channel_schema.csv in the output folder, with its SHA-256 recorded by run_manifest.json, would preserve the same source-vs-stated provenance pattern used for calibration/design.

The final resolved canonical names should also be directly available from the run manifest / RunInputs so later bricks do not reinterpret them.

Notebook

After channel count/axis roles are known:

• show one row per canonical channel index;
• prefill names from metadata only when all relevant named metadata agree;
• otherwise leave the name explicitly unset;
• let the user state/edit names;
• show where each name came from: header / stated / unnamed;
• show representative channel images alongside or near this card (see the representative-image QC issue).

Nothing is guessed from image appearance.

Consistency rules

• A stated name applies to the canonical experiment channel index, not to filenames.
• Duplicate canonical names should be refused unless we explicitly decide they are allowed.
• Empty/NA-like names should be refused with the same identity discipline as other MorphoBricks identifiers.
• Existing source metadata that contradict a stated canonical mapping must be shown explicitly; do not silently override the mismatch.
• A change to channel names should not require histogram pixels to be recomputed if channel indices/order are unchanged, but it must create a new manifest/provenance generation.
• Later bricks should read canonical names from preprocessing, not invent their own naming mechanism.

Acceptance

• Notebook can name unnamed fluorescence channels.
• Metadata names and user-stated names remain distinguishable in provenance.
• Canonical channel names are recorded in/with the preprocessing run and exposed through RunInputs.
• Reruns restore the stated names.
• Renaming a channel without changing order does not reread image pixels.
• Contradictory source metadata are never silently hidden.
• Channel names are validated as strict text identities.
• CLI/API has an equivalent way to provide/read the same channel schema.