Never crash: move the one unsafe FFI call into crates/sysmem

The macOS malloc_zone_pressure_relief call (returns freed allocator pages
after raw decodes) lived in identical copies inside the desktop app and the
CLI. It now lives in lightcraft-sysmem, an isolated L0 helper crate and the
workspace's only crate allowed unsafe: one #[allow(unsafe_code)] module, a
SAFETY comment, a safe API that returns bytes released (0 off macOS), and
tests (repeat calls, release after large frees, concurrent calls). Both apps'
alloc_release.rs now just install the safe hook, and their crate roots gain
#![forbid(unsafe_code)], so every production crate except sysmem forbids it.
AGENTS.md/CLAUDE.md name the exception, per craftrules standards/never-crash.md.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

AGENTS.md

@@ -28,6 +28,10 @@ of it. Full standard: `../craftrules/standards/never-crash.md`
   fallback (`unwrap_or…`) only where it can't silently corrupt a document. An unfinished feature returns an
   "unsupported" error or is disabled. Sole exception: a provably infallible literal, as `#[allow(clippy::expect_used)]`
   + `.expect("why it can't fail")`.
+- **`unsafe` lives only in `crates/sysmem`** (one FFI call, `malloc_zone_pressure_relief`, that returns freed
+  allocator pages to macOS after raw decodes). Every other production crate root has `#![forbid(unsafe_code)]`. A new
+  unsafe need goes in an isolated, well-tested helper crate like it: `// SAFETY:` on every block, a safe API, a safe
+  fallback where possible, and a line here naming it.
 - **Input-derived numbers are hostile:** `get()` instead of `[i]`/`[a..b]` for offsets from files, users, agents or
   arithmetic on them; checked/saturating math for lengths, offsets and counts; no division by zero, NaN/inf or negative
   casts to `usize`; cap allocations sized by input; slice strings only at char boundaries.