…python-websockets#1695)

Frame.__str__() decoded OP_TEXT frame data with a bare .decode(), which
raises UnicodeDecodeError when the frame ends in the middle of a multi-byte
UTF-8 sequence. This happens when the websockets library itself fragments a
large text message at byte boundaries (not at character boundaries) for
continuations frames (fin=False), e.g. Japanese, Chinese, or emoji text.

When DEBUG logging is enabled, the UnicodeDecodeError propagated and caused
the connection to close with code 1007 (INVALID_DATA), even though the
message was valid. The data itself was fine — only the logging was broken.

Fix: add errors='replace' to the .decode() call in Frame.__str__().
This replaces incomplete sequences with U+FFFD (replacement character),
making the log entry human-readable while never crashing the connection.

Tests: 9 new tests covering partial Japanese, partial emoji, complete frames,
ASCII, binary, and ping frames. 79 upstream tests unchanged.