Ollama gemma4:12b installed locally but not recognized by llmfit (ollama_name mapping missing for whole Gemma 4 family)

Environment

• llmfit 1.1.15
• macOS aarch64
• Apple M4 / 24 GB unified memory
• Ollama installed

Ollama

$ ollama list
gemma4:12b   4eb23ef187e2   7.6 GB   6 weeks ago

Problem

$ llmfit info "gemma4:12b"
No model found matching 'gemma4:12b'

Investigation llmfit's catalog contains Gemma 4 models (google/gemma-4-E4B-it, google/gemma-4-E2B-it, google/gemma-4-31B-it, google/gemma-4-26B-A4B-it, plus ~74 community fine-tunes matching gemma-4-12*), but grepping the full llmfit fit --json output shows "ollama_name": null on every single Gemma 4 entry. By contrast, Gemma 2 and Gemma 3 entries do carry a populated ollama_name (gemma2:9b, gemma3:12b, gemma3:27b, gemma3n:e4b, etc.).

So this isn't a fuzzy-search issue like #315 — hyphenated queries do resolve fine:

$ llmfit info "gemma-4-12b"
Multiple models match 'gemma-4-12b'. Please be more specific:
  - sakamakismile/Huihui-gemma-4-12B-it-abliterated-NVFP4A16
  - r0b0tlab/gemma-4-12B-it-nvfp4
  ...

The colon-tag form used by the official Ollama registry (gemma4:12b, and presumably gemma4:e2b, gemma4:e4b, gemma4:26b, gemma4:31b) simply has no mapping at all, so info/fit/bench/installed-detection can't resolve or recognize it, regardless of what's already pulled locally.

Expected llmfit should detect the officially distributed Ollama tags under gemma4:* as installed models (when present in ollama list) and allow info/fit/bench operations on them, the same way it already does for gemma2:* and gemma3:*.

Suggested scope Since the whole family (gemma4:e2b, e4b, 12b, 26b, 31b, plus the MLX-tagged variants) shares the same ollama_name: null gap, the fix likely belongs at the mapping-table level rather than as a single 12B-only patch.

Reference https://ollama.com/library/gemma4

Related but distinct

• #315 (closed) — catalog didn't have Gemma 4 at all; fixed by adding the models.
• #332 (closed) — missing GGUF download source mappings; fixed by #955.
• This issue is specifically about the missing ollama_name installed-detection mapping, which neither of the above touched.