Parallelize block retrieval

[tzdybal]

Currently block are retrieved from DA layer sequentially by a single goroutine.
This should be parallelized to improve sync performance during catch-up phase.

[liamsi]

We should have an experiment that tests how long it takes to sync blocks from scratch. We want to understand if using bitswap under the hood for this as well is too slow. (ref: celestiaorg/celestia-node#9 (comment))

It does not have to be through optimint but might as well. (cc @Bidon15)

[mauriceLC92]

Hey @tzdybal could you please assign this issue to me? :)