refactor(aidd-context): learn's router is a quarter of the skill, and it is always loaded

[blafourcade]

Second proof of the doctrine written in #418. Slim aidd-context:10-learn, where the router is the heaviest file, and check the contract catches a failure the split rule alone would miss.

Where the mass is

153 lines across 6 files. SKILL.md alone is 40 of them — a quarter of the skill — for four actions averaging 23 lines each. The router is denser than what it routes to.

Nothing is oversized here, so no file ceiling would ever flag it. R4 (SKILL.md <= 500 lines) passes trivially. This is a router-discipline failure: content that belongs in an action sits in the router, and the router is always read. That is why the contract must constrain the always-on surface, not file size — and learn is the case that proves it.

Scope

• Move whatever an action can own out of the router.
• Apply the contract from refactor(aidd-context): the skill template still teaches the format onboard abandoned #418; do not invent a local pattern.
• Preserve every current behavior.

Acceptance criteria

• The router carries only what R1 allows: title, chaining schema, action table, transversal rules. Zero business logic.
• The before/after lines-read-per-path is stated in the PR.
• The router is no longer the largest file of the skill.
• Nothing learn does today is lost.

[blafourcade]

Paired with #278 (let learn ingest an explicit diff or file pointer). Both rewrite 10-learn: this issue moves content out of the router, #278 adds an input path to 01-gather. Land them together — apart, the input path #278 adds would need re-slimming.