perf: trim persona system prompts — 56% token reduction

[nextlevelshit]

Summary

• Rewrote all 14 core persona .md files to lean Rules + Constraints structure
• Consolidated git forensics commands into base-protocol.md (was duplicated across 7 personas)
• Removed craftsman ↔ implementer comparison tables (pipeline authors choose personas, not agents)
• Trimmed base-protocol.md from 85 → 54 lines (removed template variable table, Known Limitations)
• Updated test thresholds: min tokens 100→60, mandatory section accepts Rules alongside Responsibilities

575 → 250 lines total (-56%), 928 deletions, 278 insertions across 31 files

Persona	Before	After	Delta
craftsman	41	15	-63%
navigator	42	14	-65%
implementer	41	15	-63%
summarizer	27	14	-48%
reviewer	33	14	-57%
philosopher	43	14	-67%
planner	35	14	-60%
debugger	36	13	-63%
auditor	34	14	-58%
supervisor	42	15	-64%
provocateur	40	14	-65%
researcher	34	14	-58%
validator	21	13	-38%
synthesizer	23	13	-43%
base-protocol	85	54	-35%

Test plan

• All existing tests pass (go test ./... — zero failures)
• TestPersonaFilesTokenRange — updated threshold, all personas within 60-400 range
• TestPersonaFilesMandatorySections — accepts Rules as mandatory section
• TestInitPersonaPromptContent — keyword assertions still pass
• .wave/personas/ synced with internal/defaults/personas/

Closes #1121