Add safe-output fail-fast mode for code push operations

[strawgate]

When push_to_pull_request_branch (or create_pull_request) fails, the safe-output pipeline still processes later outputs like add_comment.

Problem

This can post success-looking follow-up messages even though the code push failed. This causes confusion for both humans and non-humans.

Request

Add a configurable safe-output mode to run code-push outputs first and stop/skip the remaining safe outputs if a code-push output fails.

Expected behavior

• Execute code-push safe outputs first (push_to_pull_request_branch, optionally create_pull_request).
• If one fails, fail-fast the rest (or mark remaining outputs skipped with a clear reason).
• Surface this clearly in step summary/logs.

[pelikhan]

@copilot

• do not modify the ordering
• detect that one of the code push safe output failed and report it into the agent failure issue/comment in the conclusion job
• stop processing the rest of safe output with a clear cancelled message