Consistent naming for counting tokens

[sjrl]

I think it's a bit confusing how we use many different terms for number of tokens:
n_tokens, max_seq_len, max_tokens, max_tokens_limit, max_length and earlier leftover_token_len, n_full_prompt, n_full_prompt_tokens , n_skipped_tokens. Maybe we could follow a convention of using n_ when we count tokens and max_ when we set a limit? _len I would leave out. So leftover_token_len could become n_leftover_tokens and max_seq_len could become max_tokens. What do you think?

Originally posted by @julian-risch in #4179 (comment)

I agree that this is confusing and not consistent. Maybe we could combine this with your previous comment into one new PR to handle naming conventions?

Originally posted by @sjrl in #4179 (comment)

[ZanSara]

In Haystack 2.0 most of these parameters are not defined explicitly. They're mostly passed through kwargs, where I believe it makes sense to keep them exactly as their underlying backend expects them instead of normalizing them.

There is only one component that makes use of the "old" convention: ExtractiveReader. I'm going to update the name of this parameter from max_seq_len to max_tokens_per_seq and leave the others untouched.

[sjrl]

@ZanSara I think you're right that this issue is more for Haystack v1. If the ExtractiveReader is the only one doing this I'd actually to prefer to leave it as is since that is the name of the underlying variable used by HuggingFace which we decided to explicitly expose in this case instead of passing it through model_kwargs.

[ZanSara]

Ok, sounds good to me! I'll close the issue as completed then. Of course if you spot anything that I didn't catch let's open a dedicated issue and reference this one.