G-7 follow-up: backend-emitted `unavailable_reason` per signal

[zzallirog]

Context

v0.5.1 landed G-7 — dashboard tiles render explanatory N/A with hover
tooltip when a sensor is structurally absent on the host platform
(Intel iGPU has no discrete temp/power; RAPL CVE-locked; no fan tool;
etc). Reasons currently live in the frontend as a hardcoded naReasons
dict, duplicated between live-telemetry-tile.js and sparkline-tile.js.

See docs/aur-publishing.md § G-7 for the case study.

What's needed

Move reason ownership to the collector layer. Each SignalDescriptor
should optionally expose its current availability + reason at probe
time, e.g.:

@dataclass
class SignalDescriptor:
    name: str
    unit: str
    dtype: str
    ...
    unavailable_reason: str | None = None  # ← new

Then /api/discoveries already surfaces the manifest; frontend tiles
consume unavailable_reason directly instead of mapping from signal
name. Net effect: when a new platform gap is discovered, only the
collector changes — no frontend edit needed.

Acceptance

• SignalDescriptor.unavailable_reason field added with backward-compat default
• Each existing collector emits it where appropriate (rapl_energy → CVE message; intel_i915 → no power-meter; etc)
• live-telemetry-tile.js + sparkline-tile.js read from /api/discoveries instead of frontend dict
• CI guard: snapshot fixture with RAPL locked asserts the reason text round-trips end-to-end

Why this matters

Today, when an enterprise user finds a new structurally-missing signal,
the fix is split across backend (emit null) and frontend (add the reason
to the dict). Centralising in the collector turns it into a single
edit + automatic CI coverage via the platform fixtures.

Difficulty: medium. Scope: schema + 5 collectors + 2 tiles + 1 fixture
test. Estimated ~150 LOC.