Preserve double precision when auto-converting integer assignments

[fhgffy]

Summary

With auto-conversion enabled, both integer setters cast through float before writing a floating-point setting's double storage. This loses precision unnecessarily: assigning 16777217 stores 16777216, although the original value is exactly representable as a double.

Convert directly to double, matching __config_setting_get_float. The change preserves the setting type and disabled-auto-conversion behavior.

The regression covers both setters, ordinary values, positive and negative values above float's exact-integer range, boundary neighbors, and rejected writes leaving the value unchanged.

Validation

Tested on Linux x86_64 with GCC 14.2 and CMake 4.4.3, based on 1e2dedd046f5119e67b0ccbda5916e8aff55c142:

• Clean baseline: 17 native tests pass
• Regression against unchanged implementation: the new test fails eight exact-value assertions; existing tests pass
• Patched implementation: all 18 native tests pass; C/C++ libraries and examples build
• ASan/UBSan: suite passes with leak detection disabled. LeakSanitizer could not run because of the environment's ptrace restriction
• Additional local probes: C++ assignments and array elements, exact serialized output and parse-back, integer endpoints and values beyond double's exact range matching direct double conversion
• git diff --check passes

Windows, macOS, 32-bit targets, embedded hardware, and Autotools builds were not tested locally. Normal integer-to-double rounding remains for values not exactly representable as double.