feat(skills): add ARISE trace evolution, STEM pattern-to-skill, and ERL heuristics

[bug-ops]

Summary

Implements three research-backed skill evolution mechanisms in zeph-skills, addressing issues #2398, #2462, and #2463:

• ARISE (research(skills): ARISE hierarchical RL skill evolution — policy-driven selection outperforms embedding similarity across 7 benchmarks (arXiv:2603.16060) #2398): Background LLM summarization of successful multi-tool solution traces into SKILL.md description updates. Uses historical success/failure rates to re-rank embedding candidates (policy-driven selection beyond pure embedding similarity).
• STEM (research(architecture): STEM Agent multi-protocol modular design — pattern-to-skill conversion (arXiv:2603.22359) #2462): Tracks recurring tool-use patterns in skill_usage_log (migration 057). Auto-generates quarantined SKILL.md candidates when a pattern repeats >= N times with sufficient success rate — biologically-inspired skill promotion from behavior.
• ERL (research(skills): Experiential Reflective Learning — heuristic synthesis from task experiences +7.8% Gaia2 (arXiv:2603.24639) #2463): Post-task reflection extracts transferable heuristics into skill_heuristics table (migration 058). Top heuristics are injected into skill matching context as a ## Learned Heuristics section to guide future task execution.

Design

• All three features disabled by default (arise_enabled, stem_enabled, erl_enabled all false)
• All LLM calls are fire-and-forget via tokio::spawn — never block the agent response path
• All LLM-generated content starts at quarantined trust level — requires manual review before activation
• All new LLM calls expose *_provider config fields for multi-model routing
• TUI status spinners with ClearStatusOnDrop guard for all background operations
• Config migration (--migrate-config) and --init wizard updated for 14 new fields

Test plan

• cargo +nightly fmt --check passes
• cargo clippy --features full --workspace -- -D warnings passes
• cargo nextest run --features full --workspace --lib --bins — 7478/7478 pass
• New unit tests: stem.rs (9), erl.rs (8), store/skills.rs (9 new STEM/ERL queries), learning.rs (3 config tests)
• Live session test with arise_enabled = true / stem_enabled = true / erl_enabled = true in testing.toml

Closes #2398, #2462, #2463