Add if (var x = expr) and map[k].field op= rhs

[congwang-mk]

Summary

Two new statement forms for ergonomic map access, plus the doc and test work pinning them down.

• if (var x = expr) { ... } else { ... } — declaration-as-condition. Binds x only inside the truthy branch; the branch is taken iff expr produces a present value (map hit, non-null pointer return). Lowers to a single bpf_map_lookup_elem plus a null check; for struct-valued maps the body's field operations mutate the entry in place via the underlying lookup pointer (no manual write-back).
• m[k].field op= rhs — compound assignment on a struct-field of a map value. Lowers to a presence-checked if (p) { p->field = p->field op rhs; } against the same single lookup. Absent entries are a no-op (no creation), distinguishing it from the existing scalar m[k] op= rhs which creates on first use.

Bundled deliverables:

• examples/maps_demo.ks and examples/map_operations_demo.ks: rewrite the read-modify-write blocks to use the new idiom (drops 13 LoC, removes redundant write-backs).
• README.md: replace the misleading lookup_or_create snippet with paired examples; add an "Ergonomic map idioms" bullet to the solution checklist.
• SPEC.md: new §6.2.5 (compound assignment with map indexing — scalar and struct-field forms with full semantics: LHS shape, type rules, presence/creation semantics, lowering shape), new §7.5.1.2 (declaration-as-condition with single-evaluation, scoping, and lowering); assignment_statement and if_statement EBNF productions expanded from the prior shorthand into the seven forms the parser actually implements. Also fixes the §6.2.3 example, which used stats.field += 1 on a value-bound stats — a shape the parser has never accepted.
• tests/test_iflet.ml (new): 10 tests across parse, type-check, and codegen-string layers — covers struct/scalar bindings, else, else if (var ...) chains, binding-leak rejections (then-branch only), and the in-place-mutation codegen shape.
• tests/test_compound_index_assignment.ml: extend the Phase 2 test with a codegen-string check pinning the synthetic-pointer-binding path that the codegen-fix in this PR repaired (verified by temporarily reverting the fix — test flips red on "synthetic binding does NOT use deref-load init").

84 alcotest suites pass (was 83); the new test_iflet suite is the +1.

Test plan

• eval \$(opam env) && dune build && dune runtest --force — 84 suites green
• cd /tmp && cp <example>.ks . && kernelscript compile <example>.ks -o out && cd out && make — example end-to-end build for maps_demo.ks (clean) and a minimal m[k].field += 1 source (clean)
• Skim the eBPF C output for a Phase 2 source — confirms bpf_map_lookup_elem happens once, presence is checked against NULL, and the mutation is ptr->field = ...
• Spot-check generated C for a struct-map IfLet — confirms the dead value-typed local declaration plus the in-place-mutation through the lookup pointer (this is the existing user-written-IfLet shape, locked in by tests/test_iflet.ml::codegen_struct_value_binding_shape)

🤖 Generated with Claude Code