Fix varint overflow and buffer handling

Fix encodeVarint() overflow and buffer-size handling (cdf74a8)

encodeVarint() now checks for values above uint64 before writing, preventing the silent truncation that could happen with the default 10-byte buffer. It also tightens the write loop to stop at the actual buffer length and returns a clearer error when the buffer can't hold a full varint, making failures explicit instead of inconsistent.

Other misc changes

• Added regression tests for uint64 overflow with the default buffer and for undersized buffers.