HDDS-7428. Change AllocateBlock API for precise allocation and better abstraction

[kaijchen]

What changes were proposed in this pull request?

Currently, AllocateBlock takes two major arguments: size and num. There are some problems in the current API:

1. For EC replication type, the allocated size for clients to write is size * ec.numData * num, which is not a good abstraction from the HDDS client's view.
2. It assumes all blocks are in same size. However, sometimes it's useful to let SCM allocate arbitrary size the client wants (e.g. when client know the key size in advance).

In this Jira, I wish to add a requestedSize field in the AllocateBlock RPC.
SCM will calculate how many blocks to allocate based on requestedSize,
and return size of each block allocated to client.

The old size field is used for the client specified block size, if set to 0, SCM will decide the block size.
The old numBlock field is not used anymore.

What is the link to the Apache JIRA

https://issues.apache.org/jira/browse/HDDS-7428

How was this patch tested?

UT: TestSCMBlockProtocolServer#testAllocateBlock
CI: https://github.com/kaijchen/ozone/actions/runs/3345373585

[kaijchen]

Hi @JacksonYao287, this PR implements the idea of better AllocateBlock abstraction #3882 (review).

It's still work in progress though (need to add tests and handle compatibility),
can you help review the API design first?

[kaijchen]

In Ozone weekly meeting, @ChenSammi mentioned the client might need to specify custom block size for compatibility with HDFS. So I have reused the size field in proto to denote the client specified block size.

[kaijchen]

Converted to draft to address compatibility issues.