Fix aiModel json enrichment

On web ui forms, we use the aiModel json enrichment to check if there is any available model to do predictions by presenting the existing model inputs.

On the Jira demo we notice an issue, aiModel is just printing the list of inputs that all users have for that document type and we are calling ai.suggestions when all of those fields have content.

As an example we have the following models:

1. I: "file:content" > O: "tc:project"
2. I: "file:content", "tc:project" > O: "tc:components"
3. I: "file:content", "tc:project" > O: "tc:assignee"

(here is a document to this example: https://nxdev-dev-ai.beta.nuxeocloud.com/nuxeo/ui/#!/browse/default-domain/workspaces/jira/testing/NXP-3486.pdf) 
This results in the following aiModels:
aiModels: {
"inputs": [

{ "name": "tc:project", "type": "cat" }

,

{ "name": "file:content", "type": "txt" }

]
}
Nevertheless, this means the first bot is not taken into account and there is never initial suggestions to "tc:project"

Here is an expected content for aiModels for this case
aiModels: {
"inputs":[
[

{ "name":"file:content", "type":"txt" }

],
[

{ "name":"file:content", "type":"txt" }

,

{ "name":"tc:project", "type":"cat" }

],
[

{ "name":"file:content", "type":"txt" }

,

{ "name":"tc:project", "type":"cat" }

]
]
}
Of course we could do the optimization of not repeating the sets.
We also need to change the ai.suggestion call to have this new format in account.