Add DeepSeek V4 Flash model and coder composite via airouter

- models/airouter-deepseekv4flash.yaml: DeepSeek-V4-Flash on airouter.ch endpoint,
  temp 1.0 / top_p 1.0 per recommended reasoning defaults, 262k context / 65k output
- model-registry/airouter-deepseekv4flash.yaml: deepseek provider, complexity 9,
  creativity 9, cost_efficiency 10 (covered by airouter sub)
- harnesses/contexts/deepseek-code-methodology/v1: DeepSeek-specific methodology
  CLAUDE.md; same rules as qwen-code-methodology but with correct model header
- harnesses/composites/code-airouter-deepseekv4flash-repo/v1: coder composite
  using airouter/v1 context (shared endpoint + secret — no new ESO resources needed)

model-registry/airouter-deepseekv4flash.yaml

@@ -0,0 +1,15 @@
+name: airouter-deepseekv4flash
+endpoint: airouter-deepseekv4flash
+provider: deepseek
+scores:
+  complexity: 9
+  test_pass_rate: 8
+  creativity: 9
+  spec_adherence: 8
+  cost_efficiency: 10
+  context_utilisation: 9
+cost_per_1k_tokens:
+  input: 0.0
+  output: 0.0
+source: benchmark
+harness_version: null