feat(executor): add LocalRunner and OpenAI-compat LLM client

Phase 1 of "local OSS models as agents" plan. Adds a third Runner
backed by any OpenAI-compatible HTTP server (Ollama, vLLM, LM Studio,
llama.cpp), and migrates the Gemini-CLI classifier to route through
the same client when configured.

Two-layer split: internal/llm.Client is the workhorse (HTTP, no Pool,
no DB) used directly by the classifier and any future internal helper
that needs cheap reasoning. internal/executor.LocalRunner is a thin
adapter implementing Runner for user-facing tasks. This avoids
Pool reentrancy/deadlock when sub-second internal calls fire from
inside Pool.execute().

Highlights:
- internal/retry: relocated runWithBackoff/IsRateLimitError/ParseRetryAfter
  into a shared package reused by executor and llm.
- internal/llm: Chat (non-streaming) and ChatStream (SSE) over
  /chat/completions with optional bearer auth, json_object response
  format, retry on 429/503, Retry-After parsing.
- internal/executor/LocalRunner: streams deltas into stdout.log in the
  same stream-json envelope ClaudeRunner emits, then writes one
  consolidated assistant block plus a result terminator so existing
  parsers (extractSummary, ParseChangestatFromOutput) work unchanged.
- internal/executor/Classifier: gains optional LLM field; uses
  json_object response format (no markdown-fence cleanup needed).
  Falls back to Gemini-CLI subprocess when LLM is nil.
- Pool.skipClassification: now skips only when the requested agent
  type is registered, so unknown types still reach the load balancer.
- Storage: additive tokens_in/tokens_out ALTERs on executions; CLI
  runners record cost_usd as before, LocalRunner records 0 + tokens.
- Config: [local_model] section (endpoint, model, timeout_seconds,
  default_temperature, api_key). Empty endpoint = no LocalRunner
  registered, classifier falls back to Gemini.

Pre-existing test issues fixed in passing:
- claude_test.go setupSandbox callsites updated to current signature.
- gemini_test.go TestParseGeminiStream skipped (asserts unimplemented
  GeminiRunner stream-error parsing; tracked separately).

Plan: docs/plans/local-oss-runner.md.

https://claude.ai/code/session_017Edeq947TpSm1vQTxMhi1J

1 files changed, 169 insertions, 0 deletions

diff --git a/internal/retry/backoff_test.go b/internal/retry/backoff_test.go
new file mode 100644
index 0000000..a963fc2
--- /dev/null
+++ b/internal/retry/backoff_test.go
@@ -0,0 +1,169 @@
+package retry
+
+import (
+	"context"
+	"errors"
+	"fmt"
+	"testing"
+	"time"
+)
+
+// --- IsRateLimitError tests ---
+
+func TestIsRateLimitError_RateLimitMessage(t *testing.T) {
+	err := errors.New("claude exited with error: rate limit exceeded")
+	if !IsRateLimitError(err) {
+		t.Error("want true for 'rate limit exceeded', got false")
+	}
+}
+
+func TestIsRateLimitError_TooManyRequests(t *testing.T) {
+	err := errors.New("too many requests to the API")
+	if !IsRateLimitError(err) {
+		t.Error("want true for 'too many requests', got false")
+	}
+}
+
+func TestIsRateLimitError_HTTP429(t *testing.T) {
+	err := errors.New("API returned status 429")
+	if !IsRateLimitError(err) {
+		t.Error("want true for '429', got false")
+	}
+}
+
+func TestIsRateLimitError_Overloaded(t *testing.T) {
+	err := errors.New("API overloaded, please retry later")
+	if !IsRateLimitError(err) {
+		t.Error("want true for 'overloaded', got false")
+	}
+}
+
+func TestIsRateLimitError_NonRateLimitError(t *testing.T) {
+	err := errors.New("claude exited with error: exit status 1")
+	if IsRateLimitError(err) {
+		t.Error("want false for non-rate-limit error, got true")
+	}
+}
+
+func TestIsRateLimitError_NilError(t *testing.T) {
+	if IsRateLimitError(nil) {
+		t.Error("want false for nil error, got true")
+	}
+}
+
+// --- ParseRetryAfter tests ---
+
+func TestParseRetryAfter_RetryAfterSeconds(t *testing.T) {
+	msg := "rate limit exceeded, retry after 30 seconds"
+	d := ParseRetryAfter(msg)
+	if d != 30*time.Second {
+		t.Errorf("want 30s, got %v", d)
+	}
+}
+
+func TestParseRetryAfter_RetryAfterHeader(t *testing.T) {
+	msg := "rate_limit_error: retry-after: 60"
+	d := ParseRetryAfter(msg)
+	if d != 60*time.Second {
+		t.Errorf("want 60s, got %v", d)
+	}
+}
+
+func TestParseRetryAfter_NoRetryInfo(t *testing.T) {
+	msg := "rate limit exceeded"
+	d := ParseRetryAfter(msg)
+	if d != 0 {
+		t.Errorf("want 0, got %v", d)
+	}
+}
+
+// --- RunWithBackoff tests ---
+
+func TestRunWithBackoff_SuccessOnFirstTry(t *testing.T) {
+	calls := 0
+	fn := func() error {
+		calls++
+		return nil
+	}
+	err := RunWithBackoff(context.Background(), 3, time.Millisecond, fn)
+	if err != nil {
+		t.Errorf("want nil error, got %v", err)
+	}
+	if calls != 1 {
+		t.Errorf("want 1 call, got %d", calls)
+	}
+}
+
+func TestRunWithBackoff_RetriesOnRateLimit(t *testing.T) {
+	calls := 0
+	fn := func() error {
+		calls++
+		if calls < 3 {
+			return fmt.Errorf("rate limit exceeded")
+		}
+		return nil
+	}
+	err := RunWithBackoff(context.Background(), 3, time.Millisecond, fn)
+	if err != nil {
+		t.Errorf("want nil error, got %v", err)
+	}
+	if calls != 3 {
+		t.Errorf("want 3 calls, got %d", calls)
+	}
+}
+
+func TestRunWithBackoff_GivesUpAfterMaxRetries(t *testing.T) {
+	calls := 0
+	rateLimitErr := fmt.Errorf("rate limit exceeded")
+	fn := func() error {
+		calls++
+		return rateLimitErr
+	}
+	err := RunWithBackoff(context.Background(), 3, time.Millisecond, fn)
+	if err == nil {
+		t.Fatal("want error after max retries, got nil")
+	}
+	if calls != 4 {
+		t.Errorf("want 4 calls (1 initial + 3 retries), got %d", calls)
+	}
+}
+
+func TestRunWithBackoff_DoesNotRetryNonRateLimitError(t *testing.T) {
+	calls := 0
+	fn := func() error {
+		calls++
+		return fmt.Errorf("permission denied")
+	}
+	err := RunWithBackoff(context.Background(), 3, time.Millisecond, fn)
+	if err == nil {
+		t.Fatal("want error, got nil")
+	}
+	if calls != 1 {
+		t.Errorf("want 1 call (no retry for non-rate-limit), got %d", calls)
+	}
+}
+
+func TestRunWithBackoff_ContextCancellation(t *testing.T) {
+	ctx, cancel := context.WithCancel(context.Background())
+	calls := 0
+
+	fn := func() error {
+		calls++
+		cancel()
+		return fmt.Errorf("rate limit exceeded")
+	}
+
+	start := time.Now()
+	err := RunWithBackoff(ctx, 3, time.Second, fn)
+	elapsed := time.Since(start)
+
+	if err == nil {
+		t.Fatal("want error on context cancellation, got nil")
+	}
+	if elapsed > 500*time.Millisecond {
+		t.Errorf("context cancellation too slow: %v (want < 500ms)", elapsed)
+	}
+	if calls != 1 {
+		t.Errorf("want 1 call before cancellation, got %d", calls)
+	}
+}