Change test '00305_http_and_readonly' because DROP TABLE IF EXISTS now works in readonly mode in case the table doesn't exist.

Use the generic class ext::scope_guard for subscriptions in IAccessStorage instead of a special class.

Accept String or FixedString data types as key for maps in the
SummingMergeTree now that sumMap supports these two types (introduced in
bc0fbd68).
Signed-off-by: NBaudouin Giard <bgiard@bloomberg.net>

It looks like most times the original number of rows per granule is
pretty much consistent across all granules, except for the last one,
however there seems to be cases (Issue #7713) where that's not the case.

The only way I have to reproduce the crash involves using proprietary
data, and unfortunately I haven't been able to come up with a test case
that would highlight the problem - just like the person who opened issue
7713. I can't share my dataset.

Reading the code, it looks pretty obvious to me here - unless I'm
missing something, that the intention here is to copy the corresponding
granules to one another so it makes sense to index
rows_per_granule_original with the same variable as rows_per_granule.
Signed-off-by: NBaudouin Giard <bgiard@bloomberg.net>