feat(pricing): cached-input rates for gpt-5.4-mini and gpt-5.4-nano

Checked against OpenAI's pricing page: gpt-5.5 at $5 / $30 (cached $0.50)
was already right. The mini and nano models declared no cached rate, so
cached prompt tokens were billed at the full input rate.

docsgpt/core/models/openai.yaml

@@ -12,6 +12,7 @@ models:
     context_window: 1050000
     api_flavor: responses
     reasoning_effort: medium
+    # Short-context rates. Prompts over 272K tokens bill at $10 / $45 (cached $1).
     input_cost_per_million: 5.0
     output_cost_per_million: 30.0
     cached_input_cost_per_million: 0.5
@@ -20,8 +21,10 @@ models:
     description: Cost-efficient GPT-5.4-class model for high-volume coding, computer use, and subagent workloads
     input_cost_per_million: 0.75
     output_cost_per_million: 4.5
+    cached_input_cost_per_million: 0.075
   - id: gpt-5.4-nano
     display_name: GPT-5.4 Nano
     description: Cheapest GPT-5.4-class model, optimized for simple high-volume tasks where speed and cost matter most
     input_cost_per_million: 0.2
     output_cost_per_million: 1.25
+    cached_input_cost_per_million: 0.02