fix(partial-rollout): cap max_new_tokens by prior response length

[none0663]

Summary

In partial rollout, an aborted sample is resubmitted with its previously
generated tokens already attached (sample.response_length > 0). However,
sampling_params["max_new_tokens"] was still set to the full
rollout_max_response_len, allowing the engine to generate another full
budget of tokens. As a result, the total response length could exceed
rollout_max_response_len (nothing downstream clamps it).

This subtracts the already-generated length so the cumulative response stays
within rollout_max_response_len. When the budget is already exhausted,
max_new_tokens becomes 0 and the sample is marked TRUNCATED by the
existing guard.

Changes

• rollout/sglang_rollout.py (generate)
• rollout/sglang_streaming_rollout.py (generate_streaming)

Only affects runs with --partial-rollout and samples that already have
response tokens; fresh samples are unchanged.