change Block to a value struct

[gammazero]

The goal is to stop doing two allocations for each block (now it will only allocate the []byte buffer, and pass the cid.Cid, []byte pair by decomposed registers or stack).

This way of changing does not change the syntax for trivial uses of block.Block, so in theory we will have to update only producers of block.Block, not consumers.

Fixes #45

Replaces ipfs/boxo#192

Note: This change will cause a failure here:
https://github.com/ipfs/go-ipld-format/blob/0f7aff00f72e9dea0d9718bc0972e309ba7c3e8d/format.go#L27

[gammazero]

This requires many changes in boxo, which passes a traceability.Block around using a Block interface.

There are other places where more functionality is added to a block, or where something like an IPLD Node can be treated as a block, and a Block interface is to pass it.

Removing the Block interface will require many changes. Also, the same interface will reappear in other places to pass block-like objects around as blocks, or code will need to change to not use such an interface.

This PR is good for the purpose of forcing Block to be a struct value where ever possible, without having to change all code to use BasicBlock instead. For convenience, the blocks package could define an Interface type that is really the same as the Block before this PR.

[lidel]

Triage notes:

• wip
• the interface used in boxo for tracing/tests
• parking until bitswap spam reduction work is done, then it will be possible to refactor tests/tracing to remove need for interface
• tbd if we want to keep interface in addition to struct for convenience / avoid code duplication (check lotus use?)

[lidel]

Triage notes:

• turned out more involved than initially expected
• @gammazero will take one final look and write down his findings in spare time, but converting to draft for now

[gammazero]

Recommend closing this PR, See explanation.