Dedupe Verification error types (Block, Chain)

[dconnolly]

Hmm, it looks like the error type duplication is pre-existing to this PR: there's VerifyBlockError and VerifyTransactionError that both shouldn't exist. We could either fix it now, or in a later PR. The downside of fixing it later is that implementing other consensus rules requires having the right error types, so all later changes would still block on the later PR.

Comes out of #1173

1. VerifyBlockError -> BlockError
2. VerifyChainError -> (new) ChainError
3. Distinguish timeouts by wrapping Elapsed in different errors, depending on where it comes from
4. Fix the downcast from BoxError to redjubjub::Error at zebra-consensus/src/error.rs:76
   • Remove the InternalDowncastError type
   • Remove as many BoxErrors as possible

[teor2345]

Moved into the first alpha because it's causing a panic, see #1357 for details.

[conradoplg]

A related issue is discussed here. Basically the script verifier can timeout, which returns a use tower::timeout::error::Elapsed error. This is converted to a TransactionError (since that's the error type of the transaction verifier service), and since there's no matching enum item, it ends up being converted to InternalDowncastError.

I'm not sure what's the best approach for this.

• Create a Elapsed error in TransactionError? (Would solve this specific case, but there may be others)
• Try to remove the BoxError -> TransactionError conversion by using TransactionError instead of BoxError where applicable (e.g. AsyncChecks)?
• Or just leave as is?

[teor2345]

• Create a Elapsed error in TransactionError? (Would solve this specific case, but there may be others)

I think we might want a different error variant for each different kind of timeout, so we know what has timed out.

• Try to remove the BoxError -> TransactionError conversion by using TransactionError instead of BoxError where applicable (e.g. AsyncChecks)?

I'd like to replace BoxError with enum error types as much as possible.

But we'll still need it in some places, because some types and traits require BoxError (like ServiceExt::oneshot, and maybe tower::Buffer?).

[teor2345]

This isn't particularly important.