fix: tighten embedding truncation threshold (#16123)

### What problem does this PR solve?
Use a 95% max_length threshold before truncating embedding inputs, which
reduces the chance of provider-side invalid-parameter errors on
near-limit chunks.

### Type of change
- [x] Bug Fix (non-breaking change which fixes an issue)

api/db/services/llm_service.py

@@ -137,7 +137,7 @@ class LLMBundle(LLM4Tenant):
                 safe_texts.append("None")
                 continue
             token_size = num_tokens_from_string(text)
-            if token_size > self.max_length:
+            if token_size > self.max_length * 0.95:
                 target_len = int(self.max_length * 0.95)
                 safe_texts.append(text[:target_len])
             else: