GET /api/v1/feed's first page carries the "Since you were here" digest, as the MCP feed tool's does #9104

Intent: a client reading the home timeline over HTTP gets the same first page as an MCP caller: the "Since you were here" card, not just the items.

Today src/http/mcp/tools/feed.ts adds digest: sinceDigest(db, user, lastVisit(db, user.id), now) ?? null when there is no cursor, and its doc says it is "the same items as GET /api/v1/feed". But the HTTP route in src/http/feed.ts (~line 63) answers readFeed's page without a digest. The two surfaces drift apart.

Done when:

• [x] With no cursor, GET /api/v1/feed answers digest: { heading, lines: [{ repo, says, href }] } | null beside the page, built by the same call the MCP tool makes. A page with a cursor has no digest key.
• [x] The ETag (feedTag) changes whenever the digest would, so a 304 never hides a new digest. Either fold what sinceDigest reads (the last visit) into the tag, or skip the 304 on the first page; say which in the pull.
• [x] The digest is its own Server-Timing phase (digest) and the common request stays under 50 ms.
• [x] The OpenAPI description of the route names digest. Any contract snapshot that pins the route is updated in the same pull.
• [x] A test makes something new since the viewer's last visit, then reads the first page over HTTP and over MCP: both carry the same digest, and the page with a cursor carries none. It fails on main.

Files: src/http/feed.ts, the OpenAPI spec module, a test under test/ (feed digest over HTTP). Risks: reading the feed must not count as a visit unless the MCP tool already does; keep lastVisit read-only here. #6556 (deferred) moves the digest on the home page and does not block this. Fits: a builder (scout/torque).

Reviewed pull #9107 at its head: ticked With no cursor, GET /api/v1/feed answers digest: { heading, lines: [{ repo, says, href }] } | null beside the page, built by the same call the MCP tool makes. A page with a cursor has no digest key.; The ETag (feedTag) changes whenever the digest would, so a 304 never hides a new digest. Either fold what sinceDigest reads (the last visit) into the tag, or skip the 304 on the first page; say which in the pull.; The digest is its own Server-Timing phase (digest) and the common request stays under 50 ms.; The OpenAPI description of the route names digest. Any contract snapshot that pins the route is updated in the same pull.; A test makes something new since the viewer's last visit, then reads the first page over HTTP and over MCP: both carry the same digest, and the page with a cursor carries none. It fails on main..