Enhancement/issue 590 improve the fault tolerance of materialized view

[yl-lisen]

resolve #590 and close #598

Please write user-readable short description of the changes:

• Add an abstract class DB::Streaming::ISource
• Support more recovery policy RecoveryPolicy via a setting settings recovery_policy='strict' or 'best_effort':

  - Strict      /// Always recover from checkpointed
  - BestEffort  /// Attempt to recover from checkpointed and allow skipping of some data with permanent errors
  

• Add more information in error log (include the description for streaming source and sequence number (such as progressed sn, checkpointed sn, recovered sn)

If we need know what's the dirty data, we can use the following statement to query data within the specified sn range (checkpointed_sn, processed_sn]:

select * from source_stream where _tp_sn > checkpointed_sn and _tp_sn <= processed_sn
-- or
select * from source_mv where _tp_sn > checkpointed_sn and _tp_sn <= processed_sn

Usually, the data at processed_sn is the cause of the error

[chenziliang]

can we make checkpoint non virtual and remove the checkpoint, recover interface from DB::ISource?

[yl-lisen]

the checkpoint/recover virtual interface is defined in IProcessor