Drop replslot after pgstat_shutdown cause assert coredump

Hi,

When testing logical replication, I found a case which caused assert coredump on
latest HEAD. The reproduction steps are as follows:

1)
----publisher----
create table test(i int);
create publication pub for table test;
begin;

insert into test values(1);

2)
----subscriber----
create table test(i int);
create subscription sub connection 'dbname=postgres port=10000' publication pub;
- wait for a second and Ctrl-C

3)
----publisher----
commit;

I can see the walsender tried to release a not-quite-ready repliaction slot
that was created when create a subscription. But the pgstat has been shutdown
before invoking ReplicationSlotRelease().

The stack is as follows:

#2 in ExceptionalCondition (pgstat_is_initialized && !pgstat_is_shutdown)
#3 in pgstat_assert_is_up () at pgstat.c:4852
#4 in pgstat_send (msg=msg(at)entry=0x7ffe716f7470, len=len(at)entry=144) at pgstat.c:3075
#5 in pgstat_report_replslot_drop (slotname=slotname(at)entry=0x7fbcf57a3c98 "sub") at pgstat.c:1869
#6 in ReplicationSlotDropPtr (slot=0x7fbcf57a3c80) at slot.c:696
#7 in ReplicationSlotDropAcquired () at slot.c:585
#8 in ReplicationSlotRelease () at slot.c:482
#9 in ProcKill (code=<optimized out>, arg=<optimized out>) at proc.c:852
#10 in shmem_exit (code=code(at)entry=0) at ipc.c:272
#11 in proc_exit_prepare (code=code(at)entry=0) at ipc.c:194
#12 in proc_exit (code=code(at)entry=0) at ipc.c:107
#13 in ProcessRepliesIfAny () at walsender.c:1807
#14 in WalSndWaitForWal (loc=loc(at)entry=22087632) at walsender.c:1417
#15 in logical_read_xlog_page (state=0x2f8c600, targetPagePtr=22085632,
reqLen=, targetRecPtr=, cur_page=0x2f6c1e0 "\016\321\005") at walsender.c:821
#16 in ReadPageInternal (state=state(at)entry=0x2f8c600,
pageptr=pageptr(at)entry=22085632, reqLen=reqLen(at)entry=2000) at xlogreader.c:667
#17 in XLogReadRecord (state=0x2f8c600,
errormsg=errormsg(at)entry=0x7ffe716f7f98) at xlogreader.c:337
#18 in DecodingContextFindStartpoint (ctx=ctx(at)entry=0x2f8c240)
at logical.c:606
#19 in CreateReplicationSlot (cmd=cmd(at)entry=0x2f1aef0)

Is this behavior expected ?

Best regards,
Hou zhijie