fix: panic with kubectl when timestamps are enabled (#7947)

[kdeloachbayer]

Summary

Fixes a bug where using the --timestamps flag with kubectl deploy panics with:

error:panic: runtime error: slice bounds out of range

With timestamps enabled, skaffoldWriter.Write includes the length of the timestamp in its return value. But Writer implementations should return the number of bytes consumed, not written. This results in textio.PrefixWriter¹ writing N bytes to its buffer and then discarding N+len(timestamp) bytes, causing a panic.

Relevant textio.PrefixWriter source code (ref:
https://github.com/segmentio/textio/blob/master/prefix.go#L54-L55):

    c, err = w.writeLine(chunk)
    w.discard(c)

Follow-up Work

The segmentio/textio package is currently in maintenance mode. Considering it's only used in 1 file, just to append a prefix to output logs, it may be worth removing altogether. That would fix another cosmetic issue where textio calls Write twice (once for the prefix, once for the log line) which results in this output: [timestamp][prefix][timestamp][log line].

Fixes: #7947

Footnotes

1. Used by pkg/skaffold/deploy/kubectl/kubectl.go ↩

[gemini-code-assist]

Code Review

This pull request refactors the skaffoldWriter.Write method to ensure that the returned number of bytes written matches the length of the input payload, excluding any prepended timestamps. This change addresses an inconsistency where the returned value could exceed the input length. The test suite was updated to verify this behavior. Feedback was provided to handle empty input writes to prevent stray timestamps and to use more idiomatic Go patterns for string writing.