docs: design forager.media.categorize multi-faceted nested grouping

[andykais-claude]

Summary

Adds a design doc (docs/design/media-categorize.md) for a new @forager/core action, forager.media.categorize. This is a generalization of the existing forager.media.group:

• Multiple split types: tag_group, date facets (source_created_at, created_at, updated_at, last_viewed_at) bucketed by year / month / day / hour / minute, and stars.
• Arbitrarily nested splits (Cartesian product of buckets).
• Opt-in remainder bucket per split for media that doesn't match any split value.
• Same query filter surface as forager.media.search / forager.media.group.
• Deprecation plan for forager.media.group (kept around for back-compat in this PR).

No code changes — design only. Implementation will land in a follow-up PR after design approval.

Concerns called out for discussion

The doc has a dedicated Concerns & Open Questions section. Highlights:

1. Counting semantics for tag_group splits. Today's media.group counts media_reference_tag rows, which inflates counts when a media reference has multiple tags in the same group. I'm proposing categorize switches to COUNT(DISTINCT media_reference.id) since it's a new API, but want your call.
2. Date bucket value placement. The user's example put the bucket value inside by (by: { month: '2025/02' }). I'm proposing a cleaner split where by stays as the granularity and value carries the bucketed string ({ split: 'source_created_at', by: 'month', value: '2025/02' }). Calling out the deviation.
3. Multiple tag_group splits cause Cartesian blowups. Will warn at runtime but not block.
4. Pagination over deeply nested splits can produce millions of leaves with bad inputs (e.g. by: 'minute' over years of history). Paginates correctly but flagged for awareness.
5. stars granularity — currently one bucket per integer; schema is forward-compatible with optional range buckets later.
6. Web UI migration is out of scope for the implementation PR.

Testing

Design doc only — no executable changes to test. Test plan is included in the doc (10 cases covering single/nested splits, every date by, remainder semantics, query filter composition, pagination, sort orders, and categorized_media).