padl / padr: optional pad-char arg for zero-pad / dot-leader

[danieljohnmorris]

Summary

padl and padr previously hard-coded space as the pad char. That makes the
common zero-padding pattern (sortable numeric keys, fixed-column log lines)
clunky enough that personas reach for cat + manual string-building, which
costs tokens and retries. This adds an optional 3rd arg: padl s w pc /
padr s w pc, pad char defaults to space when omitted.

Manifesto framing: the cost of a missing 3rd arg here was one extra builtin
call and a few lines of glue per occurrence, multiplied by every persona who
hit it. Adding it costs two opcodes, four lines on the SPEC table, and zero
back-compat risk because the 2-arg form keeps its existing bytecode shape.

Repro

Before:

padl "42" 5         -- "   42"   (space-padded, OK)
padl "42" 5 "0"     -- ILO-T006: padl expects 2 args, got 3

After:

padl "42" 5 "0"     -- "00042"
padr "x" 4 "."      -- "x..."
padl "hi" 5 "·"     -- "···hi"     (1 Unicode scalar = 1 width unit)
padl "x" 5 "ab"     -- ILO-R009: padl pad char must be a 1-character string

What is in the diff

Six commits, one per layer so each is reviewable on its own:

1. vm new OP_PADLC / OP_PADRC (167 / 168) using the existing 2-instruction
   data-word encoding (precedent: OP_SLC, OP_CLAMP, OP_RGXSUB). Added to
   find_block_leaders. New jit_padlc / jit_padrc helpers + 5 lib unit tests
   covering happy path, multichar reject, non-string reject, Unicode scalar.
2. cranelift dispatch in both compile_cranelift.rs (AOT) and
   jit_cranelift.rs (JIT): helper FuncId, declaration, classifier list,
   dispatch + skip_next, JIT symbol registration. AOT picks the symbols up
   from libilo.a via the linker.
3. interpreter tree-walker accepts 2 or 3 args, validates pad char as a
   1-Unicode-scalar string with ILO-R009.
4. verify padl / padr joined to the arity-overload list. Third arg
   type-checked against t (ILO-T013 if it isn't).
5. test 7 new cross-engine tests covering zero-pad, dot-pad,
   already-wider, Unicode scalar, multichar reject, empty reject, 2-arg
   back-compat.
6. docs SPEC.md table + examples/pad.ilo with new -- run / -- out
   assertions (picked up by tests/examples_engines.rs). ai.txt and
   skills/ilo/SKILL.md regenerated from SPEC.md via build.rs.

Test plan

• cargo test --release --features cranelift (5215 passed, 0 failed, 45 ignored)
• cargo clippy --release --features cranelift --lib -- -W clippy::all clean
• tests/regression_pad.rs 16/16 across tree, VM, Cranelift JIT
• AOT cranelift tests pass with fresh target/release/libilo.a
• examples/pad.ilo -- run / -- out assertions pass on every engine
• 6-pass rust-review (ownership, error handling, anti-patterns, perf, unsafe, idioms) actioned: repeat_n modernisation, identical-block merge, SAFETY comment on new as_heap_ref()

Follow-ups

None. Cranelift's nil-on-bad-input behaviour for pad-char (mirrors the
existing engine-divergence on negative width) is the same gap the existing
pad_negative_width_errors test already calls out as a deferred follow-up;
harmonising the two together is out of scope for this PR.

[codecov]

Codecov Report

❌ Patch coverage is 83.09179% with 35 lines in your changes missing coverage. Please review.
✅ All tests successful. No failed tests found.

Files with missing lines	Patch %	Lines
src/vm/compile_cranelift.rs	16.66%	15 Missing ⚠️
src/vm/mod.rs	90.76%	12 Missing ⚠️
src/interpreter/mod.rs	68.00%	8 Missing ⚠️

📢 Thoughts on this report? Let us know!