Generalize inference snippets

[mishig25]

This PR:

1. Generalizes snippets so that they can support tasks that send file as an input
2. As an example, I've added snippets for img-cls task
3. If this PR gets approved, I'll add remaining tasks (such as obj-det, aud-cls, etc.) & it should close Add inference API snippets for non-NLP tasks #50 read more here

Example screenshots of img-cls task (please check whether the snippets look correct: @Narsil @osanseviero ):

[Narsil]

Snippets are good !

[mishig25]

for the file path: currently we have cats.jpg

Should it be:

1. cats.jpg
2. or ./cats.jpg
3. or {PATH_TO_IMG_FILE}
4. or something else?

[Narsil]

I think it's ok to leave cats.jpg it's pretty understandable what supposed to be there, and the error messages are also usually pretty explicit (missing file cats.jpg).

Wdyt ?

[beurkinger]

Note: you can also easily do away with the regex stuff :)

const inputsImageClassification = (_, noQuotes = false) => noQuotes ? `cats.jpgs` : `"cats.jpg"`;

const modelInputSnippets: {
	[key in PipelineType]?: (model: ModelData, noQuotes: boolean) => string;
} = {

const inputs = modelInputSnippets[model.pipeline_tag];

Not saying this is better, I'll let you judge what's the best solution.

[beurkinger]

Looks good ! I only found some admittedly petty comments to make. Still needs to try it live :)