SstMetaData's size is 0 after flush

[jiacai2050]

Describe this problem

When test #312, I find size of SstMetaData is 0 after flush memtable, this cause following error when restart(reading sst)

2022-10-19 11:49:32.342 ERRO [common_util/src/panic.rs:42] thread 'main' panicked 'Failed to create catalog manager: VisitSysCatalog { source: ReadStream { source: ErrWithSource { msg: "Read record batch", source: ReadFromSubIter { source: PullRecordBat
ch { source: DecodeRecordBatch { source: ExternalError(ParquetError(General("CachableParquetFileReader::get_metadata error: Execution error: file size of 0 is less than footer"))) } } } } } }' at "src/setup.rs:187"
   0: common_util::panic::set_panic_hook::{{closure}}

Steps to reproduce

1. Insert some data

CREATE TABLE `cpu2` (
`my_ts` timestamp NOT NULL, 
 `t1` string tag,  
 `t2` string tag,  
 `value` double,
  TIMESTAMP KEY(my_ts)) 
  with (
enable_ttl = 'false'
);

insert into cpu2  (my_ts, t1,t2, `value`) 
values
(123, 'web', 'hz', 100),
(124, 'web','hz', 101),
(123, 'ios', 'sh', 100),
(124, 'ios', 'sh', 101);

2. flush

curl --location --request POST 'localhost:5440/flush_memtable'

3. restart service

Expected behavior

No panic

Additional Information

memtable is write to sst in
https://github.com/CeresDB/ceresdb/blob/c557007e6c56390d12debe2d69478dcd73ae4796/analytic_engine/src/instance/flush_compaction.rs#L641

but we update size after this, in https://github.com/CeresDB/ceresdb/blob/c557007e6c56390d12debe2d69478dcd73ae4796/analytic_engine/src/instance/flush_compaction.rs#L657
which is too late, since it's already persisted.

[jiacai2050]

Fixed in #513