Optimize batch reconstruction in Trantor's availability module

[matejpavlovic]

The batch reconstruction currently requests a missing batch directly from all signers of the corresponding certificate. This is EXTREMELY inefficient especially at scale. E.g., with 32 nodes, at least 21 nodes sign the certificate. For a 1MB batch, a node ends up uselessly downloading 20MB of data.

Implement a smarter way of downloading the data, e.g., periodically picking a random signer of the certificate and requesting the data only from that one signer. Even better, the batch can be split into chunks and downloaded in parallel from all signers, or even erasure-coded.

https://github.com/filecoin-project/mir/blob/201930e724deee5036dfe7b4ea5051d709e8f8dd/pkg/availability/multisigcollector/internal/parts/batchreconstruction/batchreconstruction.go#L95-L99C26