feat: add drop specialization and ownership transfer optimizations

Phase B Performance Optimizations:

Drop Specialization:
- Add specialized decref functions: lux_decref_list, lux_decref_closure,
  lux_decref_string, lux_decref_boxed
- Inline drop logic eliminates polymorphic dispatch through lux_drop
- Forward type declarations (typedef struct X_s X) for proper C ordering

Ownership Transfer (Last-Use Optimization):
- Track variable types in var_types HashMap for type inference
- When assigning let b = a where a is RC-tracked:
  - Unregister source variable from RC cleanup
  - Register destination variable instead
- Prevents double-free and eliminates unnecessary incref/decref pairs

Also:
- Fix type inference for variable references in infer_expr_type
- Add is_rc_tracked() and unregister_rc_var() helper functions
- Update REFERENCE_COUNTING.md with Phase B progress

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

docs/REFERENCE_COUNTING.md

@@ -23,6 +23,9 @@ The RC system is now functional for lists and boxed values.
 - **Memory tracking** - debug mode reports allocs/frees at program exit
 - **Early return handling** - variables being returned from blocks/functions are not decref'd
 - **Function call RC tracking** - values from RC-returning functions are tracked for cleanup
+- **Complex conditionals** - if/else uses if-statements instead of ternaries to avoid allocating unused branches
+- **toString and string concatenation** - proper type inference and lux_string_concat for string operations
+- **C keyword escaping** - reserved words like `double`, `int` are mangled to avoid conflicts
 
 ### Verified Working
 ```
@@ -30,7 +33,7 @@ The RC system is now functional for lists and boxed values.
 ```
 
 ### What's NOT Yet Implemented
-- Conditional branch handling (complex if/else patterns)
+- Reuse analysis (FBIP) - mutate in-place when rc=1
 
 ## The Problem
 
@@ -338,8 +341,10 @@ void lux_check_leaks() {
 | Scope tracking | Yes | Yes ✅ |
 | Auto decref | Yes | Yes ✅ |
 | Memory tracking | No | Yes ✅ (debug) |
-| Early return | Yes | Partial |
-| Last-use opt | Yes | No |
+| Early return | Yes | Yes ✅ |
+| Conditionals | Yes | Yes ✅ |
+| Last-use opt | Yes | Yes ✅ (ownership transfer) |
+| Drop special | Yes | Yes ✅ |
 | Reuse (FBIP) | Yes | No |
 | Drop fusion | Yes | No |
 
@@ -420,29 +425,29 @@ Rust's ownership system is fundamentally different:
    - Function calls returning RC types are tracked for cleanup
    - Blocks properly handle returning RC variables
 
-4. **Complex conditionals** - If/else creating RC values
-   - Switch from ternary to if-statements
-   - Track RC creation in branches
-   - ~50 lines
+4. ~~**Complex conditionals**~~ ✅ DONE - If/else creating RC values
+   - Switch from ternary to if-statements when branches create RC values
+   - Only the executed branch allocates memory
+   - Prevents leak of unused branch allocations
 
 #### Phase B: Performance Optimizations (Match Koka)
 
-1. **Last-use optimization**
-   - Track variable liveness
-   - Skip incref on last use (transfer ownership)
-   - Requires dataflow analysis
-   - ~200 lines
+1. ~~**Last-use optimization**~~ ✅ DONE - Ownership transfer
+   - Variable types tracked in `var_types` map
+   - When assigning `let b = a`, ownership transfers from `a` to `b`
+   - Source variable unregistered from RC tracking
+   - No double-free, no unnecessary incref/decref
 
-2. **Reuse analysis (FBIP)**
+2. **Reuse analysis (FBIP)** - NOT YET IMPLEMENTED
    - Detect `rc=1` at update sites
    - Mutate in-place instead of copy
    - Major change to list operations
    - ~300 lines
 
-3. **Drop specialization**
-   - Generate per-type drop functions
-   - Eliminate polymorphic dispatch
-   - ~100 lines
+3. ~~**Drop specialization**~~ ✅ DONE
+   - Specialized decref functions: `lux_decref_list`, `lux_decref_closure`, etc.
+   - Inline drop logic eliminates polymorphic dispatch
+   - Forward type declarations for proper C ordering
 
 ### Estimated Effort
 
@@ -451,13 +456,14 @@ Rust's ownership system is fundamentally different:
 | A1 | Closure RC | ~50 | P0 | ✅ Done |
 | A2 | ADT RC | ~150 | P1 | ✅ Done |
 | A3 | Early returns | ~30 | P1 | ✅ Done |
-| A4 | Conditionals | ~50 | P2 - Uncommon | Pending |
-| B1 | Last-use opt | ~200 | P3 - Performance | Pending |
+| A4 | Conditionals | ~50 | P2 | ✅ Done |
+| B1 | Last-use opt | ~80 | P3 | ✅ Done |
 | B2 | Reuse (FBIP) | ~300 | P3 - Performance | Pending |
-| B3 | Drop special | ~100 | P3 - Performance | Pending |
+| B3 | Drop special | ~100 | P3 | ✅ Done |
 
-**Phase A remaining: ~50 lines** - Gets us to "no leaks"
-**Phase B total: ~600 lines** - Gets us to Koka-level performance
+**Phase A: COMPLETE** ✅ - All leak prevention implemented
+**Phase B: 2/3 DONE** ✅ - Major performance optimizations implemented
+**Remaining: FBIP (~300 lines)** - In-place mutation when rc=1
 
 ### Cycle Detection