feat: Implement Tier 1 Core Language Features - Enums, Options, Maps

[mjm918]

Summary

Implements the core Tier 1 language features from #15:

• Enums with variants: Stack-allocated tagged unions with unit and data variants
• Pattern matching: Destructuring in switch/case statements (case Suspended(reason):)
• Option type: Built-in some/none with is_some(), is_none(), or_default() methods
• Map type: Hash map runtime with string keys, FNV-1a hashing, linear probing

Changes

• Add Pattern AST for destructuring patterns
• Add pattern parser for switch cases
• Add enum definition tracking in Cranelift codegen
• Implement enum variant construction (unit and data variants)
• Implement enum pattern matching with destructuring bindings
• Implement option type as built-in enum
• Add hash map runtime implementation
• Implement map literals, indexing, and assignment in codegen

Test Plan

• All 116 existing tests pass
• examples/test_enum.naml - enum construction and pattern matching
• examples/test_option.naml - option type methods
• examples/test_map.naml - map operations
• examples/tier1_test.naml - comprehensive integration test

Closes #15

🤖 Generated with Claude Code

[claude]

Code Review

I've reviewed this PR and found 3 high-signal issues:

---

🚨 Critical: Use-after-free bug in resize_map

File: namlc/src/runtime/map.rs, lines 196-199

Issue: The resize_map function calls decref() on the key string before calling naml_map_set() to reinsert it. This creates a use-after-free vulnerability.

The problem:

1. Line 197 decrements the key's reference count
2. If the refcount was 1, it becomes 0 (object should be freed)
3. Line 199 calls naml_map_set which tries to use the key (hash it, incref it)

The code currently "works by accident" because it uses raw header.decref() instead of naml_string_decref(), but this is a latent bug that will manifest if anyone changes the decref call to use the proper string decref function (like the one used in naml_map_decref at line 168).

The fix: Remove the decref() call entirely. When resizing, we're just moving the same key pointer to a new location in the same map structure. The reference counting should remain unchanged - the map still owns one reference to the key.

Code reference:

naml/namlc/src/runtime/map.rs

Lines 195 to 200 in 7fa2b40

	if (*entry).occupied {
	if (*entry).key != 0 {
	(*((*entry).key as *mut NamlString)).header.decref();
	}
	naml_map_set(map, (*entry).key, (*entry).value);
	}

---

⚠️ CLAUDE.md violation: File size limits

Rule: "All source files must stay under 1000 lines. If a file approaches this limit, split it into logical submodules."
Source: .claude/claude.md

Violations:

1. namlc/src/codegen/cranelift/mod.rs - 2,691 lines (169% over limit)

   • Consider splitting into submodules like:
     • cranelift/builtins.rs (builtin functions)
     • cranelift/expressions.rs (expression codegen)
     • cranelift/statements.rs (statement codegen)
     • cranelift/types.rs (type handling)
     • cranelift/enums.rs (enum-specific codegen)

2. namlc/src/typechecker/infer.rs - 1,594 lines (59% over limit)

   • Consider splitting into submodules like:
     • infer/expressions.rs
     • infer/statements.rs
     • infer/patterns.rs

---

Summary: Found 1 critical memory safety bug and 2 code organization violations. Please address the use-after-free bug before merging.