This commit introduces fuzz testing for the `BmpTool::load()` function,
specifically targeting 54-byte BMP headers to ensure graceful failure
without heap interaction when processing malformed headers.

Key changes:
- Integrated libFuzzer into the CMake build system with an
  `ENABLE_FUZZING` option. This uses Clang and AddressSanitizer
  when enabled.
- Created a fuzzing harness (`tests/fuzz/fuzz_bitmap.cpp`) that
  feeds 54-byte data chunks from the fuzzer to `BmpTool::load()`.
- Added a GitHub Actions workflow (`.github/workflows/fuzzing.yml`)
  to automate the fuzzing process on pushes and pull requests.
  The workflow builds with Clang, ASan, and runs the fuzzer for a
  short duration.
- Added a unit test (`tests/test_bitmap.cpp`) to specifically check
  for correct error handling (`BitmapError::NotABmp`) when an
  invalid BMP magic number is encountered.

Manual analysis of `BmpTool::load()` suggests it is designed to
prevent heap allocations if header validation fails for 54-byte inputs.
The primary risks identified (integer overflows in validation,
mishandling of `ih.biSize`) are targets for the automated fuzzing
setup.

A Clang compiler crash was encountered in the development environment
when compiling `src/bitmap/bitmap.cpp` with fuzzing flags. This
prevented direct execution and iteration of the fuzzer during this
development, but the fuzzing infrastructure is provided for environments
where this compilation succeeds (e.g., the GitHub Actions runner).