docs: FFI rungs — restore libntsrocq path + #1093 orient scout

[grootstebozewolf]

Summary

• Merge complete: docs: JTS #311 in-circle design history (mesh lane session 3) #446 (JTS RelateNG touch cells: unconditional II separation + Euler EF-3 + CI upgrade #311 design history) is on main.
• FFI rungs started: restore full libntsrocq path after stale extract, and open the orientation lane for jts#1093.

Deliverables

Path	Role
docs/ffi-rungs-2026-08.md	Rung ladder 0–6, re-extract, parity YELLOW note
scripts/rebuild_oracle_ffi.sh	coqc extract → oracle_bin + libntsrocq
oracle/gen_jts1093_orient_scout.py	Ozaki vs Shewchuk Stage A vs ORIENT_EXACT
oracle/jts1093_orient_vectors.txt	10-vector table (GATE GREEN)
oracle/Makefile	jts1093-orient-scout + extract hint
docs/phase5-ffi-abi.md	§0 rebuild pointer
docs/jts-open-prs-scout-2026-08.md	Orientation lane STARTED

Local gates (WSL)

Rungs 1–2 GREEN: re-extract adds b64_orient2d_exact; make -C oracle ffi succeeds.

Rung 3 YELLOW: ffi-parity — 3 ORIENT_EXACT* failures (FFI segfault on extreme subnormal/overflow coords). Filtered/in-circle/snap/… OK (~1215 cases). Policy: exact GT via oracle_bin until fixed.

Rung 4 GREEN:

ORIENT_EXACT pin POS
Ozaki CERTAIN vs exact conflicts: 0
Shewchuk CERTAIN vs exact conflicts: 0
GATE: GREEN

Test plan

• Re-extract + make -C oracle ffi (local WSL)
• python3 oracle/gen_jts1093_orient_scout.py --oracle oracle/oracle_bin --ffi-probe oracle/ffi_probe
• Optional: fix ORIENT_EXACT FFI segfault class (follow-up)
• Optional: expand #1093 vector bank / JTS comment with numbers