feat: Ollama/gemma4 compat — /init flow, stream filter, safety fixes

provider/openai:
- Fix doubled tool call args (argsComplete flag): Ollama sends complete
  args in the first streaming chunk then repeats them as delta, causing
  doubled JSON and 400 errors in elfs
- Handle fs: prefix (gemma4 uses fs:grep instead of fs.grep)
- Add Reasoning field support for Ollama thinking output

cmd/gnoma:
- Early TTY detection so logger is created with correct destination
  before any component gets a reference to it (fixes slog WARN bleed
  into TUI textarea)

permission:
- Exempt spawn_elfs and agent tools from safety scanner: elf prompt
  text may legitimately mention .env/.ssh/credentials patterns and
  should not be blocked

tui/app:
- /init retry chain: no-tool-calls → spawn_elfs nudge → write nudge
  (ask for plain text output) → TUI fallback write from streamBuf
- looksLikeAgentsMD + extractMarkdownDoc: validate and clean fallback
  content before writing (reject refusals, strip narrative preambles)
- Collapse thinking output to 3 lines; ctrl+o to expand (live stream
  and committed messages)
- Stream-level filter for model pseudo-tool-call blocks: suppresses
  <<tool_code>>...</tool_code>> and <<function_call>>...<tool_call|>
  from entering streamBuf across chunk boundaries
- sanitizeAssistantText regex covers both block formats
- Reset streamFilterClose at every turn start

internal/router/task.go

@@ -99,17 +99,19 @@ type QualityThreshold struct {
 	Target     float64 // ideal
 }
 
+// DefaultThresholds are calibrated for M4 heuristic scores (range ~0–0.85).
+// M9 will replace these with bandit-derived values once quality data accumulates.
 var DefaultThresholds = map[TaskType]QualityThreshold{
-	TaskBoilerplate:    {0.50, 0.70, 0.80},
-	TaskGeneration:     {0.60, 0.75, 0.88},
-	TaskRefactor:       {0.65, 0.78, 0.90},
-	TaskReview:         {0.70, 0.82, 0.92},
-	TaskUnitTest:       {0.60, 0.75, 0.85},
-	TaskPlanning:       {0.75, 0.88, 0.95},
-	TaskOrchestration:  {0.80, 0.90, 0.96},
-	TaskSecurityReview: {0.88, 0.94, 0.99},
-	TaskDebug:          {0.65, 0.80, 0.90},
-	TaskExplain:        {0.55, 0.72, 0.85},
+	TaskBoilerplate:    {0.40, 0.55, 0.70}, // any capable arm works
+	TaskGeneration:     {0.45, 0.60, 0.75},
+	TaskRefactor:       {0.50, 0.65, 0.78},
+	TaskReview:         {0.55, 0.68, 0.80},
+	TaskUnitTest:       {0.45, 0.60, 0.75},
+	TaskPlanning:       {0.60, 0.72, 0.82},
+	TaskOrchestration:  {0.65, 0.75, 0.83},
+	TaskSecurityReview: {0.70, 0.78, 0.84}, // requires thinking or large context window
+	TaskDebug:          {0.50, 0.65, 0.78},
+	TaskExplain:        {0.40, 0.55, 0.72},
 }
 
 // ClassifyTask infers a TaskType from the user's prompt using keyword heuristics.