feat: Introduce pluggable VLM runtime system with preset-based configuration (#2919)

* model runtime refactoring

Signed-off-by: Michele Dolfi <dol@zurich.ibm.com>

* add test

Signed-off-by: Michele Dolfi <dol@zurich.ibm.com>

* fix code formula preset

Signed-off-by: Michele Dolfi <dol@zurich.ibm.com>

* batch prediction

Signed-off-by: Michele Dolfi <dol@zurich.ibm.com>

* use presets and new vlm options in CLI

Signed-off-by: Michele Dolfi <dol@zurich.ibm.com>

* use new model settings by default

Signed-off-by: Michele Dolfi <dol@zurich.ibm.com>

* running

Signed-off-by: Michele Dolfi <dol@zurich.ibm.com>

* update examples

Signed-off-by: Michele Dolfi <dol@zurich.ibm.com>

* fixes for running examples

Signed-off-by: Michele Dolfi <dol@zurich.ibm.com>

* keep old stage

Signed-off-by: Michele Dolfi <dol@zurich.ibm.com>

* update model

Signed-off-by: Michele Dolfi <dol@zurich.ibm.com>

* use granite 3.3 and set options

Signed-off-by: Michele Dolfi <dol@zurich.ibm.com>

* revisit init logic and propagate the proper options to the runtimes

Signed-off-by: Michele Dolfi <dol@zurich.ibm.com>

* update all stages with original setup

Signed-off-by: Michele Dolfi <dol@zurich.ibm.com>

* per stage registry

Signed-off-by: Michele Dolfi <dol@zurich.ibm.com>

* use chat template

Signed-off-by: Michele Dolfi <dol@zurich.ibm.com>

* remove duplicated predict() and factor out some utils

Signed-off-by: Michele Dolfi <dol@zurich.ibm.com>

* working picture description examples

Signed-off-by: Michele Dolfi <dol@zurich.ibm.com>

* add granite docling as code formula model

Signed-off-by: Michele Dolfi <dol@zurich.ibm.com>

* rename code formula presets

Signed-off-by: Michele Dolfi <dol@zurich.ibm.com>

* fix running minimal_vlm example

Signed-off-by: Michele Dolfi <dol@zurich.ibm.com>

* add all models to presets and run compare_vlm

Signed-off-by: Michele Dolfi <dol@zurich.ibm.com>

* remove unused repo_id

Signed-off-by: Michele Dolfi <dol@zurich.ibm.com>

* update vlm api model example

Signed-off-by: Michele Dolfi <dol@zurich.ibm.com>

* fix legacy examples

Signed-off-by: Michele Dolfi <dol@zurich.ibm.com>

* add another legacy example

Signed-off-by: Michele Dolfi <dol@zurich.ibm.com>

* fix test

Signed-off-by: Michele Dolfi <dol@zurich.ibm.com>

* avoid automatic fallback to mlx and fix end_of_utterance in codeformula

Signed-off-by: Michele Dolfi <dol@zurich.ibm.com>

* move vlm_convert_model

Signed-off-by: Michele Dolfi <dol@zurich.ibm.com>

* use new vlm runtime class

Signed-off-by: Michele Dolfi <dol@zurich.ibm.com>

* flasg for CI

Signed-off-by: Michele Dolfi <dol@zurich.ibm.com>

* rename runtimes to explicit vlm_runtimes

Signed-off-by: Michele Dolfi <dol@zurich.ibm.com>

* renaming from runtime to inference engine and model families

Signed-off-by: Michele Dolfi <dol@zurich.ibm.com>

* fixes

Signed-off-by: Michele Dolfi <dol@zurich.ibm.com>

* fix test

Signed-off-by: Michele Dolfi <dol@zurich.ibm.com>

* add docs with stages

Signed-off-by: Michele Dolfi <dol@zurich.ibm.com>

* update docs catalog page

Signed-off-by: Michele Dolfi <dol@zurich.ibm.com>

* rename runtime to inference engine

Signed-off-by: Michele Dolfi <dol@zurich.ibm.com>

---------

Signed-off-by: Michele Dolfi <dol@zurich.ibm.com>