Improve the batched background migration documentation

Problem to solve

During #378433 (closed), we heavily used the batched background migration feature.

During the implementation journey, there were several mis understandings about the system that documentation should clear:

Further details

We should have explanations on:

1. the migrations are executed sequentially, not in parallel.
2. execution is split in batches (jobs) and each batch is further split in sub batches.
3. the minimum delay between two batches is 2 minutes.
4. have some hints/guidelines on how to build the overall execution time estimation based on this comment.
5. suggest that indexes solely used by the migration should be
   1. temporary
   2. created in post migration (before the queuing migration)
6. suggest to use the database testing report to get a sense of (histograms):
   1. how much time a batch (job) takes to execute. This time should be < than the delay between batches.
   2. how much time each query is taking.

Proposal

Improve https://docs.gitlab.com/ee/development/database/batched_background_migrations.html by clearing the above points.

Who can address the issue

Anyone with a good understanding of the batched background migration feature.

Other links/references