fix(compaction): configurable summarizer tool_choice; wider overflow detection

Port of upstream grok-build 0.2.105 + 0.2.118.

- CompactionToolChoice (auto default | none), resolved env
  KIGI_COMPACTION_TOOL_CHOICE > [features].compaction_tool_choice >
  remote settings. Some OpenAI-compatible servers reject
  tool_choice:"none" with tools attached, failing every long-session
  compaction; auto keeps them working (the summarization prompt already
  forbids tool use). Threaded from agent + subagent spawn through
  CompactionConfig into generate_session_compact; the Responses gate
  counts hosted tools too, so a hosted-only request still carries the
  configured choice.
- is_context_length_error recognizes the "current message ... exceeds
  budget" phrasing, so that overflow is Fatal instead of burning the
  retry budget on a deterministic failure.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01KSy9S31pZNWXtRZNVJ1sDf

crates/codegen/kigi-sampling-types/src/error.rs

@@ -352,6 +352,7 @@ pub fn is_context_length_error(message: &str) -> bool {
         || m.contains("maximum prompt length")
         || m.contains("maximum context length")
         || m.contains("context_length_exceeded")
+        || (m.contains("current message") && m.contains("exceeds budget"))
 }
 
 /// Decide whether a [`reqwest::Error`] is worth retrying.
@@ -387,6 +388,7 @@ mod tests {
             "This model's maximum context length is 200000 tokens",
             "invalid_request_error: prompt is too long: 300000 tokens > 200000 maximum",
             "error type: context_length_exceeded",
+            "the current message exceeds budget: 120000 < 250000",
         ] {
             assert!(is_context_length_error(msg), "should match: {msg}");
         }