feat: add inject_mode (suffix/prefix) to TI pipeline

Observation: n4_baseline loss barely moved (1.025→0.965 over 3000 steps),
token_norm grew linearly without plateau — generator likely ignores last-K
CLIP positions (EOS/padding zone) where suffix injects.

Fix: add inject_mode parameter throughout the pipeline:
- "suffix": replace last K positions (original behavior, model may ignore)
- "prefix": replace positions 1:1+K right after BOS — highest attention
  weight in CLIP, much stronger gradient signal expected

Changes:
- selva_textual_inversion_trainer.py: _inject_tokens() helper centralises
  the torch.cat construction for both modes; used in training loop and eval;
  inject_mode stored in checkpoint files
- selva_textual_inversion_loader.py: reads inject_mode from checkpoint,
  includes in TEXTUAL_INVERSION bundle
- selva_sampler.py: uses _inject_tokens() via bundle's inject_mode field
- selva_ti_scheduler.py: inject_mode in _PARAM_DEFAULTS, config, and
  _train_inner call
- ti_sweep_1.json: updated with prefix_inject group (n4, n8, n4+warm);
  n4_baseline marked completed; suffix experiments retained for comparison

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

nodes/selva_sampler.py

@@ -3,6 +3,7 @@ import comfy.utils
 import comfy.model_management
 
 from .utils import SELVA_CATEGORY, get_device, get_offload_device, soft_empty_cache
+from .selva_textual_inversion_trainer import _inject_tokens
 
 
 class SelvaSampler:
@@ -118,16 +119,15 @@ class SelvaSampler:
                 neg_text_clip = feature_utils.encode_text_clip([negative_prompt]) \
                     if negative_prompt.strip() else None
 
-                # Inject textual inversion tokens into last K positions of CLIP embedding
+                # Inject textual inversion tokens into CLIP conditioning
                 if textual_inversion is not None:
-                    emb = textual_inversion["embeddings"].to(device, dtype)  # [K, 1024]
-                    K = emb.shape[0]
-                    text_clip = text_clip.clone()
-                    text_clip[:, -K:, :] = emb.unsqueeze(0)
+                    emb         = textual_inversion["embeddings"].to(device, dtype)  # [K, 1024]
+                    K           = emb.shape[0]
+                    inject_mode = textual_inversion.get("inject_mode", "suffix")
+                    text_clip   = _inject_tokens(text_clip, emb, K, inject_mode)
                     if neg_text_clip is not None:
-                        neg_text_clip = neg_text_clip.clone()
-                        neg_text_clip[:, -K:, :] = emb.unsqueeze(0)
-                    print(f"[SelVA] Textual inversion: injected {K} tokens into CLIP conditioning",
+                        neg_text_clip = _inject_tokens(neg_text_clip, emb, K, inject_mode)
+                    print(f"[SelVA] Textual inversion: {K} tokens  mode={inject_mode}",
                           flush=True)
 
                 conditions = net_generator.preprocess_conditions(clip_f, sync_f, text_clip)