feat(streaming): episode-pool iteration with decode-on-exit, video prefetch, and exact resume

Replace the shard/Backtrackable/decoded-shuffle-buffer internals with an
episode pool: each (rank x worker) consumer keeps episode_pool_size whole
episodes' tabular rows in RAM and emits uniformly random frames across
them. delta_timestamps windows become exact in-RAM slices with correct
boundary padding (the Backtrackable machinery and its lookback/lookahead
ceilings are gone), and video is decoded only when a sample is emitted,
so pool memory stays tabular-sized instead of buffer_size decoded
samples.

- Prefetch-on-admit: when streaming from a remote source, each pooled
  episode's video files download to a local cache in the background
  (refcounted, since v3 packs several episodes per file; deleted on
  eviction), so decode-on-exit reads local bytes instead of paying
  network seek latency.
- Per-consumer RNG derived from (seed, epoch, rank, worker): consumers
  decorrelated, runs reproducible, epochs reshuffle automatically.
- Deterministic fast-forward resume: load_state_dict takes the trainer's
  {batches_consumed, batch_size}; each worker re-derives its own skip
  from the DataLoader's round-robin batch assignment and replays
  tabular-only (no decode). Exact within an epoch, works with
  num_workers > 0, and the same state file serves every rank. Replaces
  the per-shard HF state_dict approach, which lived in worker processes
  and could not be captured from the trainer.
- Shard-cap default removed (max_num_shards=None uses every parquet
  shard); runtime warnings for non-divisible world sizes (datasets
  degrades to read-everything splitting) and workers left without
  shards.
- episode_pool_size replaces buffer_size (deprecated, ignored with a
  warning); decoder cache sized to the pool working set, capped at 128.

Legacy order-replication tests asserted the old buffer algorithm
step-by-step and are rewritten as behavior contracts (exactly-once
coverage, per-seed determinism, epoch reshuffle). Value-level parity
tests against the map-style dataset pass unchanged.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

slurm/benchmark_streaming_robocasa.sh

@@ -33,7 +33,7 @@ for MODE in single sarm; do
         --mode $MODE \
         --batch_size 64 \
         --num_workers 12 \
-        --buffer_size 4000 \
+        --episode_pool_size 64 \
         --num_batches 300 \
         --out_dir '"$OUT_DIR"'/node${SLURM_NODEID}
 done

slurm/run_streaming_matrix.sh

@@ -83,7 +83,7 @@ for SOURCE in $SOURCES; do
           $RUN benchmarks/streaming/benchmark_streaming.py \
             --repo_id $REPO_ID $ROOTFLAG \
             --mode $MODE --source $SOURCE --video_decode_device $DECODE \
-            --batch_size $BATCH_SIZE --num_workers $W --buffer_size $B \
+            --batch_size $BATCH_SIZE --num_workers $W --episode_pool_size $B \
             --num_batches $NUM_BATCHES --out_dir $OUT_DIR")
       jid=${jid%%;*}  # strip ';cluster' suffix on federated setups
       echo "submitted job $jid  bench_${SOURCE}_${MODE}_${DECODE}${DEPFLAG:+  (after $prev_jid)}"

slurm/train_streaming_robocasa.sh

@@ -42,7 +42,7 @@ accelerate launch \
         --repo_id '"$REPO_ID"' \
         --batch_size 64 \
         --num_workers 12 \
-        --buffer_size 4000 \
+        --episode_pool_size 64 \
         --steps 200000 \
         --save_freq 2000 \
         --log_freq 50