Using seq_length Config Rather than max_position_embeddings

Neither of these settings are used to run the model, but are referenced by external spaces like the MTEB leaderboard. Since the model doesn't technically have a max_position_embeddings due to the way ALiBi works, its a little more accurate to report a seq_length.

config.json

@@ -1,6 +1,7 @@
 {
   "_name_or_path": "hum-lodestone-v1",
   "alibi_starting_size": 4096,
+  "seq_length": 4096,
   "architectures": [
     "BertModel"
   ],
@@ -18,7 +19,6 @@
   "initializer_range": 0.02,
   "intermediate_size": 3072,
   "layer_norm_eps": 1e-12,
-  "max_position_embeddings": 4096,
   "model_type": "bert",
   "num_attention_heads": 12,
   "num_hidden_layers": 12,