Iceberg: should add caching mechanism for IcebergMetadata

[dixingxing0]

I think trino should add caching mechanism for IcebergMetadata for at least two reasons :

1. to prevent the client from using the same table multiple times in the same query and getting different results, for now iceberg connector accesses hive metastore for multiple times in a single query.
2. improve query performance, For example,when trino query a small iceberg table, we found a simple count query will cost about 4 seconds, and analyze phase cost more than 2 seconds.

[findepi]

• to prevent the client from using the same table multiple times in the same query and getting different results, for now iceberg connector accesses hive metastore for multiple times in a single query.

There should be query- (actually: transaction-) -level "cache".
For reference, in Hive this is done in TransactionalMetadataFactory.
(utilising CachingHiveMetastore under the hood).
Such "cache" should be always used, does not need to be configurable.

Furthermore, there could be additionally an opt-in longer-lived
(not transaction-scoped) metadata cache. For reference, in Hive this is
done in CachingHiveMetastoreModule (also using CachingHiveMetastore,
but with different configuration, this is a cache in a typical sense of the
word).

[phd3]

this is in progress under #7367. (not long-lived though).

[dixingxing0]

this is in progress under #7367. (not long-lived though).

Thanks @phd3, that's greate, i'm closing this issue.