fix: skip binary files in content search

[dgageot]

The search_files_content tool now sniffs file headers to detect and skip binary files before attempting to read them. This prevents slow reads, unnecessary memory usage, and potential encoding errors when encountering binary data during search operations.

The implementation detects ELF binaries, PNG images, PDF files, and other common binary formats by checking magic bytes at the start of files. Tests cover all major detection patterns and verify end-to-end behavior with mixed text and binary files in a directory.

[docker-agent]

Assessment: 🟢 APPROVE

This PR correctly implements binary file detection for search_files_content by sniffing the first 512 bytes of each file. The implementation is clean, consistent with existing patterns, and well-tested.

Approach: Checks for null bytes first (fast path for ELF, ZIP, etc.), then falls back to http.DetectContentType for magic-byte-based detection (PDF, PNG, and other formats). Silent skip on detection errors is consistent with how the rest of the walk handles I/O failures.

Test coverage: Unit tests cover empty input, plain text, UTF-8 text, ELF, PNG, PDF, and an end-to-end integration test with mixed text/binary files in a directory — adequate for the scope of change.

No bugs found in the changed code.