Fix Huffman edge cases and add CMake tests

[cynricfu]

Summary

This PR hardens the Huffman compressor/decompressor based on the earlier code review, and adds a reproducible build plus regression tests.

Fixes (priority order)

1. Huffman tree construction — Set root when only one symbol exists; reject empty input files with a clear error.
2. Single-symbol encoding — Use codeword "0" so encode/decode round-trips work for one-character files.
3. Safe byte I/O — Read input with unsigned char, reject bytes >= 128 (7-bit ASCII scope), and use binary mode consistently.
4. Modern C++ compatibility — binary_to_decimal(const std::string&), #include <cstdlib>, nullptr, and std:: in headers (no using namespace std in .h).
5. Decode robustness — Null-child checks on invalid bitstreams; fixed off-by-one bug when trimming padding on the last data byte.
6. Memory management — Destructor frees the Huffman tree and unused node_array slots.

Build & test

cmake -S . -B build -DCMAKE_CXX_COMPILER=g++
cmake --build build
ctest --test-dir build --output-on-failure

tests/run_tests.sh covers sample text round-trip, single/two-character files, DEL (0x7F), empty-file rejection, and non-ASCII rejection.

Notes

• Compressed files produced by the updated encoder remain compatible with the original on-disk layout for normal multi-symbol inputs; single-symbol files now use codeword "0" (previously broken).
• Input is explicitly limited to bytes 0–127 to match the original 128-entry symbol table.

  

[cynricfu]

This is just for trying out the cursor cloud agent.