chore(release): map islam666 for salvaged PR #39624

Ollama's default num_predict is very small (128 tokens), which causes
responses to be truncated with finish_reason='length' — especially
noticeable with Gemma4 and other models that need more output headroom.

The custom provider profile (which covers Ollama, vLLM, llamacpp, etc.)
previously had no default_max_tokens, so no max_tokens was sent in API
requests, leaving Ollama to use its very low default.

Set default_max_tokens=4096 so Ollama models produce complete responses
out of the box. Users can still override via model.max_tokens in their
config.yaml if they need a different value.

Fixes #39281

plugins/model-providers/custom/__init__.py

@@ -63,6 +63,7 @@ custom = CustomProfile(
     ),
     env_vars=(),  # No fixed key — custom endpoint
     base_url="",  # User-configured
+    default_max_tokens=4096,
 )
 
 register_provider(custom)

scripts/release.py

@@ -58,6 +58,7 @@ AUTHOR_MAP = {
     "129007007+HeLLGURD@users.noreply.github.com": "HeLLGURD",
     "290859878+synapsesx@users.noreply.github.com": "synapsesx",
     "dirtyren@users.noreply.github.com": "dirtyren",
+    "islam666@users.noreply.github.com": "islam666",
     "zhaolei.vc@bytedance.com": "zhaoleibd",
     "jeffrobodie@gmail.com": "jeffrobodie-glitch",
     "kyssta-exe@users.noreply.github.com": "kyssta-exe",