refactor(maestro-case): slim routing with progressive disclosure

[song-zhao-25]

Summary

Aggregate the approved D01-D15 Maestro Case slimming streams into one reviewable experiment. The root SKILL.md becomes a compact router while detailed guidance moves behind route-specific progressive-disclosure references. Pillar 1 remains unchanged.

Slimming metrics

Compared with merge base 5211201c2b9f732ce1b0204658ca4aafe38599da:

• Root SKILL.md: 260 → 131 lines (129 removed, 49.6% smaller)
• Root SKILL.md: 7,822 → 2,912 words (4,910 removed, 62.8% smaller)
• Full Case markdown corpus: 13,479 → 12,969 lines (510 net removed)
• Full Case markdown corpus: 159,701 → 145,126 words (14,575 net removed, 9.1%)
• Full Case markdown corpus: 1,190,218 → 1,081,335 bytes (108,883 removed, 9.1%)
• Tests: 10 paths added/updated, net +1,279 lines
• Whole PR: +2,985 / -2,216 lines, net +769 because of the added regression tests

All 20 modeled route-load scenarios became smaller; reductions range from approximately 30.5% to 62.8%.

Scope

• 47 changed paths: 37 skill-documentation paths and 10 test paths
• 10 conditional reference guides added
• Pillar 1 files remain byte-identical

Validation

• Fresh independent aggregate review converged in three cycles; final cycle had no findings
• No validated P0-P3 findings remain in the changed range
• Local tests in three parallel groups: 61 passed, 14 skipped, 0 failed
• quick_validate.py: passed
• Frontmatter and all 23 ordered critical rules: passed
• Local-link and anchor validation: passed
• Changed Python checkers compile; changed YAML tasks parse
• git diff --check: passed

Draft status

Remote coder-evals have not been launched yet. Recommended maximum-three remote matrix: connector wait, missing API workflow inline E2E, and SLA with escalation.

main advanced by three commits after the experiment base. Those commits do not overlap this PR boundary, and a synthetic merge check is clean.