fix(context): add set_skill to unbreak auto_skill_mode

[terrenceeLeung]

Allow mounting or replacing the current skill at runtime via Context.set_skill().

Summary

• Adds Context.set_skill() method and IContext.set_skill() abstract declaration so the execute engine can mount a skill after search_skill tool returns.
• Without this, enable_skill_tool=True (the default) crashes with AttributeError when the LLM invokes search_skill, because _mount_skill_from_result calls context.set_skill() which didn't exist.

Scope (One PR One Thing)

• Primary objective: Fix AttributeError crash in auto_skill_mode when LLM calls search_skill
• Out of scope: N/A
• I confirm this PR handles a single objective only: [x] Yes

Changed Files (required)

File	Reason
dare_framework/context/kernel.py	Add set_skill abstract method to IContext interface (consistent with set_tool_gateway pattern)
dare_framework/context/context.py	Add `set_skill(self, skill: Skill
docs/features/fix-context-set-skill.md	Update feature doc with merged Intent PR link

Intent / Implementation Gate (required)

• Intent PR link (docs-only): docs: intent for fix-context-set-skill #228
• Intent PR merged into main before this implementation started: [x] Yes
• Implementation PR link (this PR): fix(context): add set_skill to unbreak auto_skill_mode #227

Acceptance Criteria

• IContext declares set_skill as abstract method
• Context.set_skill() exists and sets _sys_skill
• _mount_skill_from_result no longer raises AttributeError
• Persistent skill mode (enable_skill_tool=False) is unaffected

Acceptance Pack (required)

• Contract Delta: schema — IContext gains set_skill(Skill | None) -> None abstract method; error semantics — none; retry — none
• Golden Cases: none — 4-line setter + 4-line abstract declaration, exercised by existing _mount_skill_from_result call path
• Regression Summary: existing tests pass (pytest)
• Observability and Failure Localization: none — no new events; fixes an uncaught AttributeError in execute_engine.py:244
• Structured Review Report attached: [x] Yes (below)

Test Evidence (required)

• Local commands run: pytest
• CI links or job names: PR CI
• Evidence/output summary: No regressions; the fix bridges the gap between DareAgent._mount_skill_from_result (caller) and Context/IContext (callee)

Risk and Rollback

• Risk level: Low
• Main risk points: None — additive change, no existing behavior modified
• Rollback plan: Revert commit

Structured Review Report (required)

Changed Module Boundaries / Public API

• IContext.set_skill() added as abstract method (kernel.py)
• Context.set_skill() added as concrete implementation (context.py)

New State

• New cache/global/singleton state: None
• Lifecycle and cleanup: set_skill(None) clears the skill

Concurrency / Timeout / Retry

• New concurrency points: None
• Timeout/retry locations: None
• Upper bounds: N/A

Side Effects and Idempotency

• Side effects: Mutates _sys_skill on Context instance
• Anti-duplication strategy: N/A — last-write-wins by design (one active skill at a time)

Coverage and Residual Risk

• Covered tests/evaluations: Existing execute engine path exercises this via _mount_skill_from_result
• Residual risks not covered: Custom IContext implementations must now implement set_skill (enforced by ABC)

Dependency and Lockfile Changes

• Lockfile changed in this PR: [ ] No

Agent Rules Checklist (required)

• Only task-related files changed; no opportunistic refactor
• Public interfaces/data structures unchanged unless explicitly required
• Tests added/updated and evidence attached
• Any skip/only/exclude usage is explained and reviewed — N/A
• Merge will be done by approved reviewer (no self-merge auto-ship)