feat: show OME channel names in the LayerPanel + let users rename channels

[aelefebv]

What this does

Two related channel-identity features:

1. Display channel names from the OME metadata (generic). lucida now reads
   attributes.ome.omero.channels[].label from a dataset's zarr.json and shows it in the
   LayerPanel instead of the positional Ch N. It's keyed on the standard OME-Zarr omero
   block — no vendor/format-specific code — so any OME-Zarr with omero labels benefits
   (CZI/ND2/LIF alike). Reading the cheap omero JSON, not the bundled OME-XML, keeps the
   dataset-open path fast and off the XML-parsing surface.

2. User channel rename. A new SetChannelName ViewportCommand carries a per-channel
   ChannelSettings.name override. It mirrors the existing SetChannelColormap path exactly,
   so the rename persists in saved views and broadcasts to followers like every other
   per-channel setting. An inline rename input (gated on the same canEdit permission as the
   layer rename) sits next to each channel label; an empty rename clears the override.

Display precedence: user override → omero label → Ch N, applied at every channel-naming
site (label, eye-button aria-label, colormap selector, contrast control).

How it's threaded

• lucida-store/src/parse.rs — parse the omero channels (label + optional color), fail-soft on
  malformed/partial omero (interior gaps backfilled positionally, never mislabels).
• lucida-content/src/image.rs — MultiscaleInfo.channel_infos: Vec<ChannelInfo>
  (#[serde(default, skip_serializing_if]).
• lucida-store/src/import.rs — populate it on import (single image + plate).
• lucida-core — SetChannelName command + ChannelSettings.name.
• web — mirror the types, thread the names into the LayerPanel, add the rename UI.

Back-compat

Additive and snapshot-safe: new Rust fields are #[serde(default)], the web field is optional
with a Ch N fallback, ChannelSettings/colormaps/saved-view encoding are otherwise unchanged
(colormaps stay positional in this PR — this changes names, not colors), and channels still
bind by index. Datasets without an omero block (incl. all existing lucida-written stores) keep
showing Ch N.

Tests + verification

cargo test --workspace, clippy -D warnings, fmt, wasm-pack build, and the web
tsc/eslint/pnpm test (1497) all pass. New coverage: omero→manifest e2e + back-compat,
the api-wiring fail-safe (non-plate scene_root-raises, channel-count mismatch), the
projection/precedence, and the rename command (round-trip, apply, clear, selection-epoch).

Verified live against a real 7-channel Leica STELLARIS confocal store (enriched by the
companion zarrmony change): the LayerPanel shows DAPI (405 nm), ALEXA 594 (590 nm),
ALEXA 750 (753 nm), ALEXA 488 (499 nm), ALEXA 647-R-PE (653 nm), ALEXA 555 (553 nm),
ALEXA 700 (696 nm); renaming a channel overrides + persists, clearing falls back to the omero
label.

Related

The companion converter change that writes these omero labels for Leica confocal LIFs is
ferrinm/zarrmony#32 (which in turn ports the fix upstream to bioio-devs/bioio-lif#52).