Avoid etcd request too large

[lmatz]

etcd request too large error appeared in #6726 and #7315

Correct me if I am wrong,
the etcd request size can be linearly related to

1. the number of queries (during recovery, migration only)
2. the parallelism of each query
3. the size of the metadata of each query

This error now also appears in one POC case: only one query, but the query is particularly long and complicated

Of course, we can increase the --max-request-size of etcd to avoid the error as usual.
Although the doc of etcd said that a request of larger size may cause longer latency(I suppose it can be superlinear otherwise it is trivial), the etcd latency should be negligible compared to the total e2e latency of creating MV or recovery.

But in long term, we probably need to

1. compress the part that can be compressed,
2. or come up with a new and more concise representation.

Not particularly urgent as --max-request-size can be adjusted, but open this issue as a record