v1.10.0

Headline: worst-case shadow-stack bound — the AbsInt StackAnalyzer analogue
for Wasm (FEAT-021 slice-1). scry now computes a sound over-approximation of
a component's linear-memory shadow-stack usage, reusing the call graph + Tarjan
SCCs it already builds. This is also the first release to publish the
scry-sai-* libraries to crates.io (the publishing setup below activates on
this v* tag).

Added — FEAT-021 slice-1 (worst-case shadow-stack bound, DD-016)

• StackUsage on the analysis result: per-function frame sizes + a
  max_stack_bytes bound. The shadow stack is the C-style stack Rust/LLVM keep
  in linear memory via the __stack_pointer global; each function's frame is
  the constant its prologue (global.get $sp; i32.const F; i32.sub; global.set $sp) subtracts, and the worst case is the deepest weighted path
  through the call graph — exactly AbsInt StackAnalyzer's value-analysis →
  call-graph → longest-path shape.
• Reuses the FEAT-006/007 backbone: a frame-size detector feeds
  compute_stack_usage, which folds frame(f) + max over callees callees-first
  over the existing Tarjan-SCC reverse-topological order. New global-section
  parsing identifies the mutable-i32 __stack_pointer.
• Soundness guardrails (DD-016), never an under-count: a recursion SCC →
  Unbounded (no finite bound without a ranking function); a dynamic
  alloca / unrecognised prologue / ambiguous SP global → Unknown; a module
  with no mutable-i32 global → Bytes(0) (no shadow stack); host/WASI frames
  out of scope (guest shadow-stack only). Unbounded/Unknown are never
  treated as zero.
• Mechanized soundness (proofs/rocq/StackBound.v, stack_bound_sound):
  for any active call chain (each step a static callee, each frame ≤ the
  detected size), the frame-sum is ≤ the analyzer's per-function bound — so the
  reported max over all chains over-approximates the true peak under an
  operational push/pop semantics. No admits/axioms; rocq-proofs CI job (and
  locally with Coq 9.0.1).
• Fixtures + oracles: fixture-12-stack-chain (frames 16/32/8 → 56 bytes),
  fixture-13-stack-recursion (→ Unbounded), fixture-14-stack-dynamic
  (→ Unknown), all in the live scry-mcdc corpus. MC/DC proved 180 → 204
  (mac local; CI/linux gate). SCRY_VERSION → 1.10.0.

Falsification statement

What v1.10 claims: for a non-recursive component using the standard
__stack_pointer prologue, the reported max_stack_bytes ≥ the true peak
shadow-stack usage on every concrete run. Falsifier: a .wat whose true peak
(measured by instrumenting the SP global in wasmtime) exceeds the reported
bound. Does not claim: a bound for recursion (reported Unbounded), dynamic
frames (Unknown), or host/WASI stack (out of scope) — slice-2 surfaces the
result in WIT + a host-test oracle; slice-3 adds call_indirect target-set
precision. FEAT-021 stays proposed until those land.

Added — crates.io publishing (activates on this tag)

• The reusable library crates are now publishable to crates.io, mirroring
  pulseengine/synth: a Publish to crates.io workflow
  (.github/workflows/publish-to-crates-io.yml) fires on every v* tag
  alongside release.yml, driven by scripts/publish.rs — a leaf-first
  dependency-ordered publisher with a 10-attempt / 40s retry loop to ride out
  crates.io index propagation, using the org-wide CRATES_IO_TOKEN.
• Published set (the genuine reusable work body), under the scry-sai-*
  namespace (SAI = Sound Abstract Interpretation): scry-sai-interval,
  scry-sai-taint, scry-sai-octagon, scry-sai-provenance (leaves) →
  scry-sai-core. The crates.io package name is scry-sai-*; the Rust
  crate ([lib] name, e.g. scry_interval) and the on-disk directory
  (crates/scry-interval/) keep their existing names — the witness DEC-034
  trick — so use scry_interval and the Bazel target paths are unchanged.
  Internal path deps now carry version so cargo publish rewrites them to the
  crates.io coordinate.
• Wasm-component crates renamed into the namespace: wasm-lattice →
  scry-sai-lattice, scry-analyzer → scry-sai-analyzer ([package] name
  only — [lib] name, directories, and Bazel targets unchanged). They remain
  publish = false and ship as signed .wasm release assets: they are cdylib
  Wasm components built from Bazel-generated WIT bindings, so a host
  cargo publish cannot build them and a cargo add consumer cannot use a
  cdylib component (the witness precedent — publish the libraries, ship the
  component). Making them genuine crates.io libraries would require a
  wit_bindgen::generate! binding refactor that touches the shipped //:scry
  composition for no consumer benefit; deferred. The scry-host-tests and
  scry-mcdc harnesses remain unpublished.

Changed

• workspace.package.version aligned to the release-tag line: 0.1.0 →
  1.9.0 so the crate version on crates.io matches the v1.x release
  artifacts and SCRY_VERSION. A future release bump must move the tag, this
  version, and the internal path-dep version fields in lockstep (the
  crates.io publish workflow asserts tag == workspace version). The first
  crates.io publish therefore lands on the next release tag.