fix: add vLLM/local server error patterns + MCP initial connection retry (#9281)

Port two improvements inspired by Kilo-Org/kilocode analysis:

1. Error classifier: add context overflow patterns for vLLM, Ollama,
   and llama.cpp/llama-server. These local inference servers return
   different error formats than cloud providers (e.g., 'exceeds the
   max_model_len', 'context length exceeded', 'slot context'). Without
   these patterns, context overflow errors from local servers are
   misclassified as format errors, causing infinite retries instead
   of triggering compression.

2. MCP initial connection retry: previously, if the very first
   connection attempt to an MCP server failed (e.g., transient DNS
   blip at startup), the server was permanently marked as failed with
   no retry. Post-connect reconnection had 5 retries with exponential
   backoff, but initial connection had zero. Now initial connections
   retry up to 3 times with backoff before giving up, matching the
   resilience of post-connect reconnection.
   (Inspired by Kilo Code's MCP server disappearing fix in v1.3.3)

Tests: 6 new error classifier tests, 4 new MCP retry tests, 1
updated existing test. All 276 affected tests pass.

agent/error_classifier.py

@@ -156,6 +156,18 @@ _CONTEXT_OVERFLOW_PATTERNS = [
     "prompt exceeds max length",
     "max_tokens",
     "maximum number of tokens",
+    # vLLM / local inference server patterns
+    "exceeds the max_model_len",
+    "max_model_len",
+    "prompt length",             # "engine prompt length X exceeds"
+    "input is too long",
+    "maximum model length",
+    # Ollama patterns
+    "context length exceeded",
+    "truncating input",
+    # llama.cpp / llama-server patterns
+    "slot context",              # "slot context: N tokens, prompt N tokens"
+    "n_ctx_slot",
     # Chinese error messages (some providers return these)
     "超过最大长度",
     "上下文长度",