Keep the reference platform a continuous target: detect parity gaps and default drift rather than assuming agreement

> 🤖 Generated by the Agentic Engineer

Part of #6875.

## Problem

#6875 names `devantler-tech/platform` as the **continuous** reference target — not a one-time
inventory. The reference platform keeps moving: it gains components, changes defaults, and hardens
configuration as its own roadmap advances. KSail's built-in platform will silently drift away from it,
and the drift is invisible precisely because both sides look healthy on their own.

That is the same failure shape already recorded elsewhere in this portfolio: a check that reads clean
because nothing is comparing the two things, not because they agree.

Without this child, #6877's inventory is accurate on the day it is written and decays from then on,
and the claim "KSail can provision a platform like the reference one" quietly stops being true with
nobody noticing.

## What this child produces

A repeatable comparison, run on a cadence rather than by hand, that answers:

- Which capabilities the reference platform has that KSail cannot provision — the **parity gap**.
- Which KSail slot defaults have diverged from the reference platform's current choice — **default
drift** — including version pins.
- Which capabilities KSail provisions that the reference platform has since dropped — the reverse
case, which is evidence a default is stale.

Each difference is either a tracked issue or a recorded, justified divergence. **A difference that is
neither is the defect this child exists to prevent.**

## Explicitly not

This does not make KSail a clone of one operator's platform. Divergence is legitimate and expected —
what is not legitimate is *undetected* divergence. The output is a decision prompt, not an
auto-sync.

## Acceptance criteria

- [ ] A check enumerates the reference platform's current capabilities and compares them against
KSail's slot inventory and defaults.
- [ ] It distinguishes parity gap, default drift, and reverse drift, and reports version pins.
- [ ] It fails loudly on an unreviewed difference and is quiet on a recorded, justified one.
- [ ] It runs on a cadence without a human starting it, and a stale or non-producing check is itself
detectable — an absent result must never read as agreement.
- [ ] Its findings enter the normal issue queue rather than a report nobody reads.

## Note on the failure mode this guards

The check must not be able to pass by comparing nothing. Whatever form it takes, it needs a control
proving it can still see both sides — an empty comparison is a claim about the comparison, not about
the platforms.

## Size

Medium.

The issue names no files, tests, or implementation entry point. Start by reading #6875 and #6877, then inspect the reference platform and KSail's slot inventory; done means a scheduled comparison reports parity gaps and drift, rejects unreviewed differences, and produces an issue-queue result.

Written by the indexing model from the issue text.