quantize: prevent input/output file collision

[Anri-Lombard]

Summary

• Add check to prevent quantizing when input and output files are the same
• Uses std::filesystem::equivalent() to handle path variations (e.g., ./file vs file)
• Prevents file corruption caused by mmap reading from a file being simultaneously written

Test

# Before fix: file gets corrupted (19MB -> 727KB)
llama-quantize model.gguf model.gguf Q4_0

# After fix: clean error message, file preserved
main: error: input and output files are the same: 'model.gguf'

Fixes #12753

[Anri-Lombard]

Updated to use the error_code overload as suggested. Thanks @danbev