MUL-6669 feat(inbox): add From and unread-only filters (#7537)

* feat(inbox): add From and unread-only filters

Two dimensions join the existing status/priority filters, both from data
the Inbox row already carries — no backend change.

From matches the whole per-issue group rather than the one row that
survives deduplication: an actor belongs to a NOTIFICATION, so matching
only the surviving row would hide an issue Alice commented on the moment
someone else touches it. `inboxActorIndex` carries the rest of the group
alongside the deduplicated list, built from the same rows the matching
deduplication consumed.

Unread reads the surviving row's own `read`, which is what the unread
badge counts — so the filter and the number next to Inbox cannot
disagree. Unlike the issue dimensions it keeps notifications without an
issue, the quick-create failures most worth screening for.

`inboxFiltersForPrioritySupport` now preserves the dimensions it is not
degrading, and the workspace entry is dropped only when every dimension
is empty.

Co-authored-by: multica-agent <github@multica.ai>

* refactor(inbox): scope the From filter to the row on screen

From matched any actor in a notification's per-issue group, so selecting
Alice could return a row rendering Bob's avatar and "Bob commented" — the
filter answered "issues Alice has ever touched" while the row it produced
said something else.

Match the actor of the notification the list actually renders instead, so
what you filter by is what you see. This is what the label, the avatar and
the detail line already imply, and it removes the group index, its two
`queries` wrappers and the third parameter of filterInboxItems.

"Issues Alice participated in" remains a real but different question; it
needs its own name and its own affordance, not this one.

Co-authored-by: multica-agent <github@multica.ai>

---------

Co-authored-by: J <bohan@devv.ai>
Co-authored-by: multica-agent <github@multica.ai>