Fix & improve docs related to story augmentation

[samsucik]

What problem are you trying to solve?

The augmentation CLI arg is quite mysterious even though it impacts Core training a lot. We should add more info to the docs, as discussed in #6550.

What's your suggested solution?

Changes we should make:

• Make it explicit that augmentation is a crucial parameter when trying to speed up Core training, but less augmentation can make TED's performance worse
• Correct the docs to state the correct default augmentation factor (they say it's 20 but it's 50).
• Make it explicit that augmentation affects only TED (out of the policies we provide)

Optionally, we might add these recommendations based on our observations:

• If your config contains TEDPolicy and no memoization policies, start with story augmentation turned on (e.g. the default value of 50)
• Otherwise (if a memoization policy is used too), try out also with story augmentation turned off -- to potentially speed up training

Examples (if relevant)

No response

Is anything blocking this from being implemented? (if relevant)

No response

Definition of Done

• Discussed and decided whether or not to add the recommendations to the docs
• Updated the docs

[samsucik]

Btw this issue has a natural followup here, but that one's more technical and has been decided to be done only after 3.0.