Fix MoE expert traffic, surface search limits, retire dennard naming (#2)

feat: audit popular model profiles (#1)

fix: charge gemma nvfp4 fp8 kv cache

docs: specify compare table metrics

feat: audit deepseek v4 flash profile

fix: use exact deepseek ds4 gguf profile

fix: byte-audit qwen3.6 nvfp4 profile

feat: add audited deepseek r1 profile

feat: add qwen3.5 mixed dtype profiles

feat: add multimodal text decode profiles

docs: keep workload settings out of profiles

docs: use self-contained model profiles

docs: specify profiled bounds coverage targets

docs: require per-model binding coverage

docs: expand profiled bounds specifications

docs: add profiled bounds implementation plan