[Feature] Proposal for data api

linked pr is #1 but was suggested discussion may be easier in an issue.

# Implementation of data api proposal

Based on the atlas API v2.

## Dataset Structure

```text
proposal/example_dataset/
└── data-volumes/
├── 550e8400-e29b-41d4-a716-446655440000/ # GFP channel
│ └── 1_0/
│ ├── metadata.json
│ └── volume.ome.zarr/
│ └── .placeholder
└── 660e8400-e29b-41d4-a716-446655440001/ # autofluorescence channel
└── 1_0/
├── metadata.json
└── volume.ome.zarr/
└── .placeholder
```
### Notes

- each dataset is given a UUID
- volume data stored as zarr
- We use fields from openminds, stored in a json
- Metadata will be stored as a sqlite database so we can query quickly. We will build the database from the jsons.

## Metadata fields

These are not from openMINDS (or I couldnt find them):

* `id` - UUID, unique dataset identifier.
* `injection_coordinate` - Specific coordinate in the related coordinate space (e.g. `[6600, 4000, 5400]`).

## openMINDS-derived fields
* `name` - dataset name.
* `contributors` - Labs or authors who produced the data.
* `description` - What the dataset contains.
* `dataset_version` - Semantic version (e.g. `"1.0"`).
* `license` - Usage terms (e.g. `"CC-BY-4.0"`).
* `species` - species name (e.g. "Mus musculus") - I guess this should follow the brainglobe atlas api since each dataset should have a corresponding CCF in the atlas api
* `developmental_stage` - Life cycle class: adolescent, adult, embryo, infant, juvenile, etc.
* `injection_target` - Injection target regions, nested with the annotation set they belong to: `{regions, annotation_set: {name, version}}`. The `name` is a BrainGlobe atlas name (e.g. `"allen_mouse"`).
* `technique` - Method of accomplishing a desired aim. 194 valid values from openMINDS.
* `measured_quantity` - What the voxel values represent (e.g. fluorescence intensity, cell density). Since each channel is its own dataset, this is always a single quantity.
* `studied_target` - What were we trying to measure (e.g. DRD1, C-Fos, Nissl )
* `digital_identifier` - DOI for the associated publication (e.g. `"10.1234/example"`).
* `anatomical_axes_orientation` - BrainGlobe orientation code (e.g. `"asr"`).
* `voxel_size_um` - Voxel size in micrometers.
* `coordinate_space` - The BrainGlobe atlas name the data is registered to (e.g. `"allen_mouse"`). Must be a valid atlas in `brainglobe_atlasapi`. Resolution is not included here — it is captured by `voxel_size_um`.
* `shape` - Volume dimensions in voxels.

## my fields for handling population average datasets
* `sample_number` - number of animals used to create this average.
* `number_of_female` - the number of female animals used to create this average.
* `number_of_male` - the number of male animals used to create this average.
* `number_of_hemaphrodite` - the number of hemaprodite animals used to create this average.
* `age` - The average age of animals used in this dataset
* `age_units` - the units for interpreting the age variable

## Channel Splitting

Each channel of a multi-channel volume is stored as a **separate dataset** with its own UUID and metadata. This means:

- A single-channel acquisition produces one dataset.
- A multi-channel acquisition (e.g. two fluorescence channels) produces one dataset per channel.
- Related channels can be linked via a shared `subject_id` (animal UUID) so they can be grouped back together when needed.

This keeps each dataset simple (one volume = one measured quantity) and avoids the complexity of per-channel metadata arrays.

## Additional Metadata Fields for Channel Linking

* `subject_id` - UUID identifying the animal (subject) the data came from. All datasets derived from the same animal share this ID, allowing multi-channel or multi-modal data to be linked.
* `project_id` - UUID identifying the project this dataset belongs to. Datasets produced as part of the same set (e.g. a study or publication) share this ID.
* `channel_name` - Human-readable name for this channel (e.g. `"GFP"`, `"tdTomato"`, `"autofluorescence"`).

Start by reviewing the linked PR #1 and the proposal's dataset structure, including metadata.json, volume.ome.zarr, and the SQLite database plan. Compare the openMINDS-derived and channel-linking fields with the existing Atlas API v2 approach. Done requires an agreed proposal and an implementation scope, but the issue does not define which changes remain after the linked PR.

Written by the indexing model from the issue text.