fix: use caller's class name as default model_type for subclasses

[AtharvaJaiswal005]

Summary

• Use self.__class__.__name__ instead of hardcoded "SentenceTransformer" as the default model_type in _get_model_type()
• Fixes subclassed SentenceTransformer loading older models (without model_type in config) with wrong pooling defaults

Root Cause

_get_model_type() hardcoded the fallback to "SentenceTransformer" in two places:

1. When no config_sentence_transformers.json exists
2. When the config exists but has no model_type key

Meanwhile, _model_config["model_type"] is set to self.__class__.__name__ (e.g. "MySentenceTransformer"). This mismatch caused the comparison at line 325 to fail, falling through to _load_auto_model instead of _load_sbert_model, resulting in wrong pooling defaults (mean pooling instead of CLS token).

Reproduction

from sentence_transformers import SentenceTransformer

class MySentenceTransformer(SentenceTransformer): ...

model = MySentenceTransformer("BAAI/bge-small-en-v1.5")
# Before fix: "No sentence-transformers model found... Creating a new one with mean pooling."
# After fix: loads correctly with CLS token pooling

Changes

• sentence_transformers/SentenceTransformer.py: Changed default from "SentenceTransformer" to self.__class__.__name__ in both fallback paths of _get_model_type()

Fixes #3536