fix: subpage create conflict

[AlexHardAtWork]

Fixes #24.

What

• siblings() no longer filters archived_at IS NULL.
• Both create paths map DocStoreError::Conflict to 409 doc.sort_key_conflict instead of
  falling through to a 500.

Why

Archived documents keep their sort_key and still count towards
UNIQUE (workspace_id, parent_id, sort_key), but siblings() hid them from key generation.
So after deleting a subpage, the next one regenerated the same key and the INSERT collided.
Letting generation see exactly what the constraint enforces removes the collision.

The 409 is worth having regardless: move_to already returns that code for the same
conflict, so the doc-mutating endpoints now agree, and a conflict stops being reported as an
internal error.

Why no migration

The alternative is a partial unique index over live rows only. That needs a migration and
a change to restore(), since a restored document's key may have been reused while it was
archived. The one-line query change achieves the same invariant with less surface.

Nothing user-visible changes: all three siblings() callers use it purely to compute a
sort key, never for display, and list_alive() still filters archived documents from the
tree.

Tests

New store test covering delete-then-create (crates/knot-storage/tests/documents.rs);
make test green. The 409 arm is deliberately untested — with generation fixed, only a race
between concurrent creates can still reach it, and there is no mock DocStore in the tree.

Not included

The lexorank exhaustion from the "bonus" section of #24 (the 8th subpage under one parent
regenerates "0m") is a separate defect and left for its own PR.