t27f: why card -- a social card that draws one deps_why chain, box per spec #10716

Boundary

New file specs/t27f/why_card.t27, a new skill card specs/skills/t27-why-card.t27 with its seal, and a row in .claude/skills/t27f/SKILL.md. Nothing else changes. The module reads the chain that deps_why in specs/t27f/spec_deps.t27 (t27#10705, PR #10715) writes, and returns an ffmpeg filter graph for a 1080x1350 card. It posts nothing. Publishing stays the owner's call.

User Scenarios

1. TRI CLAW is asked why lesson_reel pulls in ci/affected. It pipes t27c use-edges into why_card with WHY_A and WHY_B set, renders the graph with ffmpeg, and gets a card: one box per spec from A down to B, each pair joined by a uses connector, a hop count in the footer.
2. A chain that does not exist, a refused walk, a broken chain or one too long for the card returns no graph, and the reason is printed, so no card is ever drawn that does not match a real deps_why answer.

Requirements

1. FR-001: card_graph MUST draw exactly one box per spec of the chain, in chain order, and one connector per edge.
2. FR-002: every box MUST use the exact style of spec_graph_view (stroke VIEW_LABEL_STROKE[0], fill VIEW_LABEL_FILL[0]), because every edge deps_why writes is a resolved use line.
3. FR-003: names MUST drop a leading specs/ and a trailing .t27, and the text size MUST fit the box at the font's 3/5 em advance.
4. FR-004: for every chain length from 2 to 7 boxes, the column MUST sit inside the band 260..1150 px, centred, with gaps of at least 30 px.
5. FR-005: card_graph MUST return 0 for none, refused, 0 hops, a chain whose hops do not join A to B, a non-exact edge, a name outside [A-Za-z0-9_/.-] or longer than 58 bytes, and a chain of more than 7 specs.
6. FR-006: a graph that does not fit the caller's buffer MUST return cap + 1, never a truncated graph.
7. FR-007: each rule MUST have a test, and each refusal a negative control.

Success Criteria

• t27c test-report specs/t27f/why_card.t27 reports every test passing, with 0 vacuous.
• A lab render of a real corpus chain has, at the sampled pixels, the box fill inside every box and the backdrop in every gap. A control pixel outside the column reads as backdrop.
• The seal verifies and the corpus ratchet stays CLEAN.