Slai Agent Runtime：Improve Linux PTY output draining for large command logs

[Slai.life (huong8373tt-beep)]

Summary

Slai Agent Runtime encountered a performance bottleneck while running Linux container workloads through RepoLaunch: large PTY transcripts were repeatedly reprocessed as each output chunk arrived.

This change makes PTY-output draining incremental:

• Stop re-normalizing and regex-scanning the full accumulated transcript for every read.
• Detect the PS1 completion delimiter incrementally, retaining only the suffix needed to handle delimiters split across reads.
• Parse command metadata once, after the completion delimiter is observed.

Problem

LinuxRuntime._read_raw_output appended each new output chunk and then rescanned the entire transcript. With large build or test logs, this produced quadratic host-side work. The command inside the container could finish successfully while the evaluator was still processing terminal output and had not yet advanced to its result artifacts.

Validation

• Added regression coverage for large PTY output and completion delimiters split across reads.

• Verified the targeted output-drain tests with:

  python -m unittest tests/linux_output_drain_test.py

Context

Developed and validated by Slai Agent Runtime during Linux container-based evaluation workloads.