Fix precision in fixed-pt conversion to avoid overflow

The previous implementation used ((angle * TWIN_ANGLE_360) >> 15) which
caused integer overflow when angle values reached their maximum range of
[-9092, 9092]. The multiplication 9092 * 4096 results in 37,240,832,
exceeding the 16-bit signed integer limit.

This commit simplifies the expression by factoring out the constants:
- Original: angle * TWIN_ANGLE_360 / 32768
- Since TWIN_ANGLE_360 = 2^12 and 32768 = 2^15
- Simplified: angle * 2^12 / 2^15 = angle / 8

Use integer division (not right shift) to match original truncate-towards-zero
behavior for negative values. Right shift would round towards negative infinity,
causing precision errors for ~87.5% of negative angle values (all non-multiples
of 8).

Benefits:
1. Eliminates integer overflow risk
2. Avoids incorrect double-precision cast
3. Removes floating-point operations
4. Preserves exact precision for all angle values

Addresses review comments from @jouae and @weihsinyeh in PR #110.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <[email protected]>

Author: jserv

src/trig.c

@@ -148,7 +148,14 @@ static twin_angle_t twin_atan2_first_quadrant(twin_fixed_t y, twin_fixed_t x)
         }
     }
 
-    return (twin_angle_t) (double) angle / (32768.0) * TWIN_ANGLE_360;
+    /* Fixed-point conversion: angle / 32768 * TWIN_ANGLE_360
+     * Simplified: angle * 2^12 / 2^15 = angle / 8
+     * Use integer division (not right shift) to match original truncate-towards-zero
+     * behavior for negative values. Right shift would round towards negative infinity,
+     * causing precision errors for non-multiples of 8.
+     * Avoids overflow since no intermediate multiplication required.
+     */
+    return (twin_angle_t) (angle / 8);
 }
 
 twin_angle_t twin_atan2(twin_fixed_t y, twin_fixed_t x)