Make pipelinerun/taskrun logs agent-friendly

Summary

tkn pipelinerun logs and tkn taskrun logs are likely the most-used commands by AI agents debugging CI/CD failures. The current output is optimized for human consumption with color codes and interleaved multi-container output that's hard to parse programmatically.

Proposal

• --output json for structured log output with metadata (task name, step name, timestamp, message)
• Ensure --follow works well in non-interactive/agent contexts
• Clear separation of log lines per step/container in structured mode
• Consider --timestamps in a parseable format (RFC3339)

Example

tkn pipelinerun logs my-run --output json

{"task": "build", "step": "build-step", "timestamp": "2026-05-12T10:00:00Z", "message": "Building image..."}
{"task": "build", "step": "build-step", "timestamp": "2026-05-12T10:00:01Z", "message": "Successfully built abc123"}

NDJSON (one line per log entry) is the natural format here since logs are a stream.

/kind feature

Start with the implementations of tkn pipelinerun logs and tkn taskrun logs, then trace how log lines are collected, interleaved, colored, and followed. Define the structured NDJSON fields and parseable timestamp behavior from the proposal, including non-interactive --follow; done means both commands support agent-friendly structured output with clear task and step separation.