builtins: add ord and chr for per-char codepoint round-trip

[danieljohnmorris]

Summary

Adds ord c:t -> n (first char's Unicode codepoint) and chr n:n -> t (codepoint to single-char string). Closes a real-world gap surfaced in the assessment doc: every persona doing per-character text work (NLP tokenisation, classifier features, ASCII-fold lowercasing) had to hand-roll a 26-entry mset lookup map because num "H" returns a parse error rather than 72. Two opcodes for what would otherwise be ~30 lines of boilerplate per program is a clean manifesto win.

Standard arity for both, table-stakes builtins that every language ships. Pattern-matched against PR #190 (upr/lwr/cap) for full coverage across tree walker, bytecode VM, and Cranelift (AOT + JIT).

Repro

Before:

$ echo 'f>n;num "H"' | ilo /dev/stdin --run-tree f
# returns Err("H") — no way to get 72 without a hand-rolled 26-entry map

After:

$ echo 'f>n;ord "H"' | ilo /dev/stdin --run-tree f
72
$ echo 'f>t;chr 72' | ilo /dev/stdin --run-tree f
H

What's in the diff

• builtins: register ord and chr in name table and type sigs — enum slots, name resolver, parser arity (1-arg both), verifier signatures (ord is t->n, chr is n->t).
• interpreter: dispatch ord and chr in the tree walker — validates non-empty string for ord; validates finite non-negative integer codepoint within u32 / char::from_u32 range for chr.
• vm: add OP_ORD=153 and OP_CHR=154 with VM dispatch — same validation in VM arms. Adds jit_ord / jit_chr extern-C helpers (return TAG_NIL on invalid args matching at/hd/padl precedent the JIT can't raise typed errors without unwinding).
• cranelift: wire ord and chr through AOT and JIT backends — helper FuncIds, declarations, emission arms in both compile_cranelift.rs and jit_cranelift.rs. Register-type classifier learns OP_ORD writes a number and OP_CHR writes a non-number.
• tests: cross-engine regression coverage for ord and chr — 12 cross-engine tests + an engine-split empty-string error test (tree+VM raise; Cranelift returns nil). Covers ASCII letters/digits, multi-byte UTF-8 (é = U+00E9 = 233), codepoint 0 round-trip, takes-first-char-only, chr(ord(c)) and ord(chr(n)) round-trips. Plus examples/ord-chr.ilo with -- run / -- out directives so the example-engine harness exercises it.

Test plan

• cargo build --features cranelift clean
• cargo test --release --features cranelift --test regression_ord_chr — 13/13 pass
• cargo test --release --features cranelift --test examples_engines — example exercised across all engines
• cargo test --release --features cranelift — full suite green
• cargo clippy --release --features cranelift --all-targets -- -D warnings — clean
• cargo fmt clean
• CI green on push

Follow-ups

None planned. If real-world use surfaces a need for ord/chr over a list (e.g. map ord (spl s "") for full string-to-codepoints), the existing map builtin already covers that path.

[codecov]

Codecov Report

❌ Patch coverage is 52.60116% with 82 lines in your changes missing coverage. Please review.
✅ All tests successful. No failed tests found.

Files with missing lines	Patch %	Lines
src/vm/mod.rs	47.69%	34 Missing ⚠️
src/verify.rs	40.00%	18 Missing ⚠️
src/interpreter/mod.rs	52.94%	16 Missing ⚠️
src/vm/compile_cranelift.rs	17.64%	14 Missing ⚠️

📢 Thoughts on this report? Let us know!