[on hold] Add `dataset.tags` and expose search feature

[tcompa]

Here is a recap of open discussion items:

1. About using tags for sorting datasets: [on hold] Add dataset.tags and expose search feature #3217 (comment)
2. About tag scope: [on hold] Add dataset.tags and expose search feature #3217 (comment)
3. About tag name normalization: [on hold] Add dataset.tags and expose search feature #3217 (comment)

(EDIT: I removed the original detailed checklist, since this will need to start over after the discussion moves forward)

[jluethi]

Simple case, because just arbitrary user tags
(no defaults or presets like for tasks & templates)

[tcompa]

A small specification question:

When adding a tag to a dataset (be it upon dataset creation, or when editing an existing dataset), we want to let the user either type the tag or pick it from a list of existing tags.
Question: should the list of existing tags be project-specific or should it include all tags of all datasets that the user has access to?

cc @jluethi

[jluethi]

Question: should the list of existing tags be project-specific or should it include all tags of all datasets that the user has access to?

I'd say tags to pick from should be all tags the user has set in any of the projects the user has access to. If we can be smart in ordering them, that would help once users have many tags.

Some ordering ideas: tags from the same project first and after that most recently used first (tbd what definition of used we'd want here. Can be most recently added or sorting by dataset creation time or by dataset used in workflow => is one of those reasonably accessible to us?)

[jluethi]

Another dataset tag idea if sorting would get tricky: We could have a user page (somewhere) on dataset tags. It could have 2 functionalities:

1. Manage your tags, e.g. determine the order in which they will show up, rename a tag etc.
2. Find datasets based on tags (i.e. show me all the "favorite dataset" tagged datasets across projects)

[tcompa]

Can be most recently added or sorting by dataset creation time or by dataset used in workflow => is one of those reasonably accessible to us?

In the current version we keep track of creation timestamp for datasets, but not of the last-edited timestamp.

We should review how such a database-query would look like (where the potential complexity is that we need to simultaneously order by dataset timestamp and also deduplicate). cc @ychiucco could you have a look?

Another dataset tag idea if sorting would get tricky: We could have a user page (somewhere) on dataset tags. It could have 2 functionalities:

1. Manage your tags, e.g. determine the order in which they will show up, rename a tag etc.
2. Find datasets based on tags (i.e. show me all the "favorite dataset" tagged datasets across projects)

This kind of feature is obviously doable, but it differs from the one above - since it also introduces a per-user available-tags concept (as opposed to only introducing tags at the per-dataset level).

We should review (probably live) what the goal is. Some examples:

1. Per-dataset tags:
   • Tags are only associated to a single dataset
   • Homogeneity across datasets is not guaranteed
   • Homogeneity across datasets is made easier through suggestions of existing tags
2. Global tag system
   • A list of tags exists at the per-user level
   • In principle this list could be somewhat integrated with tags used for other objects (projects, tasks, templates) - depending on how we plan the corresponding features.
   • The user can act on this list, e.g. by removing a tag or determining a specific order.
   • Bulk features across projects and datasets may be introduced. Examples are "Show me the list of all datasets marked as nice", or "Delete all datasets marked as test". These are both possible also with per-dataset tags, but they benefit from a single list of tags.
   • [likely other features, TBD]

[tcompa]

Motivated by the discussion above, but partly unrelated: if we pick a capitalization rule, we can likely reduce the number of test vs Test discrepancies.

Some options:

1. All lower case (nice datasets, testing, hackaton 2025, hcs, ...)
2. Each word is capitalized (Nice Datasets, Testing, Hackaton 2025, Hcs, ..)