Zig FFI: 0.15.2 migration debt — 11 compile errors in ffi/zig/src/

[hyperpolymath]

Surfaced by #31's smoke gate + lmdb fix. Pre-existing main breakage (all 3 most recent main runs failed at Zig FFI Build & Test).

After the lmdb-dev + Zig 0.15 quick-fixes in #31 (financial_extract / gpu_ocr / hw_crypto.zig:117 inline-asm), the compile reaches deeper into ffi/zig/src/ and surfaces 11 more 0.15.2 incompatibilities:

Category A — visibility (1 site)

• src/docudactyl_ffi.zig:371 — gpu_ocr.ddac_gpu_ocr_submit not marked pub. Fix: add pub to declaration in src/gpu_ocr.zig.

Category B — *const DynLib cast-discards-const (4 sites)

• src/gpu_ocr.zig:169, 184, 199 — lib.lookup(...) requires *DynLib, gets *const DynLib.
• src/ml_inference.zig:117 — lib.close() same problem.
• Fix: refactor receivers to take mutable *DynLib OR @constCast at call sites. The std.DynLib API moved to mutable-receiver in Zig 0.15.

Category C — std.Target API change (1 site)

• src/hw_crypto.zig:65 — std.Target.x86.featureSet(target.cpu.features) — featureSet signature changed to expect []const Feature not Feature.Set.
• Fix: pass features differently or use the new Target.Cpu.Feature.Set API.

Category D — net stream API change (1 site)

• src/dragonfly.zig:46 — stream.handle.setReadTimeout(...) — stream.handle is now i32 (raw FD), not a struct with methods.
• Fix: use std.posix.setsockopt(stream.handle, std.posix.SOL.SOCKET, std.posix.SO.RCVTIMEO, ...) or equivalent.

Category E — additional unused/var-const strictness (likely several more)

Once A-D are fixed, expect a second wave of unused capture / var-never-mutated / expected pub errors per the Zig 0.15 strictness regime.

Why this isn't in #31

#31 (Wave-3 Chapel rehab) ships the docudactyl-side fixes (chpl 2.8.0 + smoke + CI). The FFI is a separate subsystem; this migration debt has been latent on main since the Zig 0.15 toolchain bump (~2026-05-26 per snifs#30 / [[reference_zig_015_breaking_changes]]). Fixing it properly needs:

1. Domain knowledge of GPU OCR + ml inference + dragonfly socket code paths
2. Possibly toolchain matrix in CI (0.14 + 0.15 transition)
3. A separate review pass (not bundled with Chapel work)

Recommended: stand up a focused PR(s) per category, with build-ffi gated locally on each pass.

Owner action — confirm whether this should:

• (a) be tackled before the next docudactyl release
• (b) be delegated to a Zig-focused contributor or sub-agent fanout
• (c) inform a broader estate-wide Zig 0.14 → 0.15 migration tracker (the same shape may apply to echidna / other repos with Zig FFI)