Fix two out-of-bounds read issues when handling truncated UTF-8 input

[frankslin]

Two independent out-of-bounds read issues were identified in OpenCC's UTF-8 processing logic when handling malformed or truncated UTF-8 sequences.

1. MaxMatchSegmentation:
   NextCharLength() could return a value larger than the remaining input size.
   The previous logic subtracted this value from a size_t length counter,
   potentially causing underflow and subsequent out-of-bounds reads.

2. Conversion:
   Similar length handling could allow reads past the end of the input buffer
   during dictionary matching, potentially propagating unintended bytes to the
   conversion output.

This patch fixes both issues by:

• Explicitly tracking the end of the input buffer
• Recomputing remaining length on each iteration
• Clamping matched character and key lengths to the remaining buffer size
• Preventing reads past the null terminator

The changes preserve existing behavior for valid UTF-8 input and add test coverage for truncated UTF-8 sequences.

These issues may have security implications when processing untrusted input and are classified as heap out-of-bounds reads (CWE-125).

[frankslin]

Fixes #997