M3: Ranges (infix <-, inclusive, 4<-1 descends)

[assapir]

M3: Ranges via infix <-

Adds an infix range operator <- that materializes an inclusive []Num:

• 1 <- 4 → [1, 2, 3, 4]
• 4 <- 1 → [4, 3, 2, 1] (descends when the left end is larger)
• 5 <- 5 → [5]

It is pure array sugar — there is no distinct Range type. The result is a []Num, so it composes with .size, indexing [i], and for loops.

Implementation (localized to the new precedence level + a checker arm + range codegen)

• Lexer: reuses the existing LeftArrow token — no new token.
• Parser: new non-associative parse_range level between comparison and pipeline. The for header still consumes its own <- in parse_for_loop, so for n <- coll parses exactly as before (regression-tested).
• AST: new Expr::Range { start, end, span } (+ span() arm).
• Type checker: Num <- Num → []Num.
• Codegen: generate_range emits the shared {ptr, size} array shape with a runtime fill loop; element count |hi - lo| + 1 and ascending/descending direction are decided at runtime (ends may be dynamic). Backing storage is GC-allocated (__alloc) so the array may safely escape the frame.

Coexistence

The pre-existing for n <- collection => body loop is unchanged. The infix range only applies in general expression position (between two value expressions); the for header's <- is parsed separately. Both forms are covered by tests.

Tests & docs

• examples/ranges.ql (exit code 14), wired into tests/examples_test.rs — runs under both JIT and native AOT (clang + gcc).
• tests/ranges_test.rs: (1<-4).size == 4, single-point, ascending values, descending order, dynamic ends, range drives a for loop, and for n <- [...] still runs.
• Parser unit tests: infix <- parses as Expr::Range; for header still parses as ForLoop.
• LANGUAGE.md: symbol table, prose section, and feature matrix.

Gate

cargo build, cargo test (incl. native-AOT examples gate with clang+gcc), cargo fmt --check, cargo clippy --all-targets -- -D warnings — all green. Ran /code-review and /simplify; addressed findings (reused ptr_len_struct_type, collapsed a redundant subtraction, trimmed docs/tests). The doc-comment overstatement flagged by review was corrected; numeric-edge findings (f64 truncation, NaN/overflow) were left as-is for consistency with the language's unguarded f64-everywhere 0.9 core.

🤖 Generated with Claude Code