`fix(gkd): defer full-vocabulary local teacher logits to forward_step`

Checklist / 检查清单

• I have searched existing issues, and this is a new bug report. / 我已经搜索过现有的 issues，确认这是一个新的 bug report。

Bug Description / Bug 描述

For a separate local teacher with exact full-vocabulary logits
(gkd_logits_topk is None), _compute_teacher_logits_local eagerly materializes
one [S, V] teacher logits tensor per rollout micro-batch at batch
preparation, and all of them stay alive for the whole generation cycle. With
64 queries × n=4 (256 rollouts), up to 256 full [S, 151k] logits tensors are
retained at once, OOMing long teacher responses (observed in practice).

How to Reproduce / 如何复现

Change

• Split the per-batch eager loop into _compute_teacher_output_local, a
  single-micro-batch helper that forwards one local teacher microbatch and
  returns its TeacherOutput.
• _compute_teacher_logits skips the eager materialization when
  gkd_logits_topk is None (full-vocabulary mode): the encoded batches keep
  teacher_model_inputs instead of teacher_output.
• forward_step computes the teacher output just in time for the micro-batch it
  is about to train on, so only one [S, V] logits tensor is alive at a
  time. Fails closed with RuntimeError if a batch has neither an eager
  teacher_output nor usable teacher_model_inputs.
• Compressed top-k logprobs ([S, K]) remain eager — they are small; the
  deferral targets the exact full-vocabulary path only.
• Self-distillation behavior is unchanged (still recomputed per train step via
  _on_train_step_batch).

Additional Information / 补充信息

No response

Locate _compute_teacher_logits_local, _compute_teacher_output_local, _compute_teacher_logits, and forward_step, then trace how encoded batches carry teacher data. Verify that full-vocabulary local-teacher logits are computed only for the micro-batch being trained, while compressed top-k outputs remain eager and self-distillation is unchanged. Done means the full-vocabulary path retains only one logits tensor and fails clearly when neither teacher output nor inputs are available.