feat(models): OpenRouter model picker + AJAX refresh + cost preview (v0.11.0 complete)

- AJAX endpoint 'prautoblogger_refresh_models' for manual model registry refresh
- Admin UI button on AI Models tab to trigger refresh (nonce + capability gated)
- Cost preview integration in field renderer using historical token averages
- New public method on Cost Tracker: get_avg_tokens_for_stages(array $stages, int $days = 30): array
- Enhanced field renderer with capability-specific cost calculation (outline+draft+polish for writing, single stage for analysis/editor)
- Autoloader updated to scan ajax/ directory for model registry refresh handler
- Admin JS enhanced with AJAX toggle handler for refresh button
- Template updated with refresh button + nonce in AI Models section
- CHANGELOG + CONVENTIONS + version bump to 0.11.0
- All files under 300-line cap, PHPCS-clean

Spec: https://github.com/peptiderepo/prautoblogger/issues/111
Open questions all resolved per seq-02 plan-ack.

includes/core/class-cost-tracker.php

@@ -210,4 +210,61 @@ class PRAutoBlogger_Cost_Tracker {
 	public function get_current_run_cost(): float {
 		return $this->current_run_cost;
 	}
+
+	/**
+	 * Get average input/output token counts for given stages over a time period.
+	 *
+	 * Used by the model picker field renderer to calculate estimated costs
+	 * per generation based on historical token usage. Maps setting IDs to their
+	 * constituting stages (e.g., writing model controls outline + draft + polish).
+	 *
+	 * @param array<string> $stages Stage names ('analysis', 'outline', 'draft', 'polish', 'review').
+	 * @param int           $days   Historical window (default 30 days).
+	 *
+	 * @return array{avg_prompt_tokens: float, avg_completion_tokens: float, sample_size: int}
+	 *         Returns empty counters if no history. Never throws.
+	 */
+	public function get_avg_tokens_for_stages( array $stages, int $days = 30 ): array {
+		global $wpdb;
+
+		if ( empty( $stages ) ) {
+			return [
+				'avg_prompt_tokens'      => 0.0,
+				'avg_completion_tokens'  => 0.0,
+				'sample_size'            => 0,
+			];
+		}
+
+		$cutoff_time   = time() - ( $days * DAY_IN_SECONDS );
+		$stage_list    = implode( ',', array_map( array( $wpdb, 'prepare' ), array_fill( 0, count( $stages ), '%s' ), $stages ) );
+		$table_name    = $wpdb->prefix . 'prautoblogger_generation_log';
+
+		// phpcs:ignore WordPress.DB.PreparedSQL.InterpolatedNotPrepared -- Prepared stage list via array_map + prepare
+		$query = $wpdb->prepare(
+			"SELECT AVG(input_tokens) as avg_input,
+                    AVG(output_tokens) as avg_output,
+                    COUNT(*) as sample_count
+             FROM $table_name
+             WHERE stage IN ( $stage_list )
+             AND timestamp >= %d",
+			$cutoff_time
+		);
+
+		// phpcs:ignore WordPress.DB.PreparedSQL.NotPrepared -- phpcs pragma above
+		$result = $wpdb->get_row( $query, ARRAY_A );
+
+		if ( ! $result || 0 === (int) ( $result['sample_count'] ?? 0 ) ) {
+			return [
+				'avg_prompt_tokens'      => 0.0,
+				'avg_completion_tokens'  => 0.0,
+				'sample_size'            => 0,
+			];
+		}
+
+		return [
+			'avg_prompt_tokens'      => (float) ( $result['avg_input'] ?? 0 ),
+			'avg_completion_tokens'  => (float) ( $result['avg_output'] ?? 0 ),
+			'sample_size'            => (int) ( $result['sample_count'] ?? 0 ),
+		];
+	}
 }