feat(remove-outputs): tag-scoped retention for limit and age

[timdegroot1996]

Implements #309

Make limit and age respect a tags filter so housekeeping can be scoped to a subset of runs instead of acting globally:

• limit + tags -> keep the N newest runs carrying any of the given tag(s), remove older matching runs, leave untagged runs untouched.
• age + tags -> remove only tagged runs within the age range, leave untagged runs untouched (supports both older 10d and younger -10d).

CLI combination syntax uses the regular comma separator, e.g. -r "limit=10,tag=nightly" / -r "age=10d,tag=nightly,tag=prod". arguments.py detects the combination, validates it, and rewrites it into the internal limit=N;tag=... / age=X;tag=... form (the legacy ; form still works). Rules:

• tags may combine with EITHER limit OR age, not both (error).
• only one limit/age may be combined with tags (error).
• a clear INFO line explains exactly what will be deleted; when independent options (index/alias/run_start) are also present the deletion order is printed.

Server /remove-outputs scopes the same way when tags is sent with limit and/or age, and rejects the limit+age+tags combination before removing anything.

Backward compatible: plain limit/age/tag (no partner) behave exactly as before; the elif dispatch in database.py is reordered so combined queries route to the scoped handler.

Docs (CLI + server), OpenAPI examples, and the admin UI labels updated. Also documents the previously under-documented negative-age (-10d) behavior in the docs.

[timdegroot1996]

Pushed a follow-up commit that narrows scope: dropped the `age` + `tags` combination.

This PR's title/description advertise both `limit`+`age` scoping, but issue #309 only asked for `limit`+`tags`. Keeping both meant three places (arguments.py's CLI comma-detection, server.py's JSON-body handling, database.py's dispatch) each had to know about a hand-rolled `limit=N;tag=x`/`age=X;tag=x` mini-DSL, plus a three-way `limit`+`age`+`tags` ambiguity error that only existed because of the extra combination.

What changed:

• `arguments.py` — `_process_remove_runs` only detects/rewrites `limit`+`tag(s)`; `age`+`tag(s)` now passes through untouched (two independent ops, same as `main`). Dropped the limit+age+tags ambiguity check (exit 3) since it's no longer reachable.
• `database.py` — `_remove_by_age` reverted to its pre-PR signature (no `run_tags`/tag filtering). `_remove_by_limit` unchanged. Dispatch order: `limit=` still checked before `tag=` (required — the combo string contains the substring `tag=`), `age=` moved back after `tag=` since it no longer needs to precede it.
• `server.py` — dropped the age-scoping branch and the `limit`+`age`+`tags` rejection guard; `scope_tags` is now `tags and limit` only.
• Docs (`basic-command-line-interface-cli.md`, `dashboard-server.md`, `admin.html`) and tests trimmed to match — removed the age+tag examples/table rows/test cases, kept everything limit+tag.

Net diff vs. the previous branch state: -138 lines. `limit`+`tags` behavior, examples, and tests are unchanged — this only removes the unrequested `age`+`tags` half. Title/description of this PR should probably be updated to drop the "and age" framing before merge. Full python suite (311 tests) green; CLI help-text fixture updated and reverified.