refactor(jit): simplification and cleanup

[fffonion]

Summary

• Removes the disconnected src/vm/native/inline.rs emitter and keeps the active native lowering path explicit.
• Replaces repeated JIT builtin specializations with declarative, typed builtin specs covering:
  • length, type, and predicate helpers;
  • string/bytes transforms and fallible regex replacement;
  • collection queries, mutations, and iterator operations;
  • effect and failure-exit metadata consumed directly by lowering.
• Preserves typed callable-frame shapes, metadata-reserved root locals, and AOT preemption for oversized monolithic lowering.
• Improves wide WAF JIT/AOT execution while retaining default rule-set semantics and native trace diagnostics.
• Moves JIT diagnostic assembly into its own module without changing the public snapshot or dump API.
• Parameterizes Plain, Inherited, and Owned tail-call ABI lowering while retaining separate runtime entries, direct-link behavior, W^X buffers, and keepalive ownership.
• Places the active implementation plans in the repository plans/ directory and adds the slot-state/deopt model proposal.

Compatibility and scope

• No runtime tail wrappers were merged across incompatible ABIs.
• The slot-state document is a design proposal only; deopt and region-fusion behavior remain unchanged.
• The optional dynasm-rs link spike is outside this PR.

Validation

• cargo fmt --all -- --check
• cargo test --workspace --all-features
• cargo clippy --workspace --all-targets --all-features -- -D warnings
• cargo doc --workspace --all-features --no-deps
• cargo check --workspace --no-default-features
• AArch64 workspace check and strict Clippy
• wasm target check
• Focused native side-link, owned-transfer, inherited-handoff, diagnostics, and direct-link regression tests

Performance note

The tail-wrapper refactor keeps ABI selection in lowering time, with no runtime ABI dispatch added. Repeated direct-link test timing did not show a regression signal; its process-level measurements are retained only as a coarse guard, not as a throughput claim.