homepage: add per-model input modalities to the LLM cards

Each LLM card now lists its accepted inputs as a line (Inputs: text · image ·
audio), sourced from the harness models.yml:
- frank-large: text (no vision projector)
- frank-medium: text, image (Gemma4 vision path)
- frank-small: text, image, audio (Gemma4 projector audio encoder; audio is
  not in models.yml's input list only because OMP's schema rejects it, but the
  endpoint accepts input_audio parts)

New .desc-inputs CSS + bare-mode reset; /docs model guide gains an Inputs line
per model. bats 30/30 + live render verified.

shared/homepage/services.json

@@ -270,6 +270,7 @@
     "details": {
       "role": "Deep reasoning model",
       "spec": "Qwen3.8-27B (UD-Q8_K_XL, 31.5 GiB) — 262k-token context, one conversation at a time, text only.",
+      "inputs": ["text"],
       "use": "Cybersecurity and debugging. Requires Frank's permission."
     },
     "url": "",
@@ -286,6 +287,7 @@
     "details": {
       "role": "Primary model",
       "spec": "Qwen3.8-27B-Uncensored (MXFP4/W4A8, 19.4 GiB) — 262k-token context across up to eight conversations; sees images. Beats Claude Sonnet 5 per Artificial Analysis.",
+      "inputs": ["text", "image"],
       "use": "Software engineering, system administration, and instruction tasks."
     },
     "url": "",
@@ -302,6 +304,7 @@
     "details": {
       "role": "Fast utility model",
       "spec": "Gemma4-12B QAT (Q4_K_M, 7.4 GiB) — 1M-token pool across four conversations; fastest at images.",
+      "inputs": ["text", "image", "audio"],
       "use": "Agentic search, RAG, research subagents, translation, roleplay, and summarization — any small task that doesn't need much intelligence."
     },
     "url": "",