fix(nodestore): Avoid full bigtable scans in get_multi

[kcons]

Add an id_list deduplication so that we can correctly recognize cache hits and avoid full table scans in bigtable.
Add a test that demonstrates the fix.

We observed some timeouts in nodestore.get_multi calls, with lookups for a small number of IDs taking 40s+.
One commonality was that they were lookups that included duplicated IDs and cache hits; this is because we
recognize full cache hits by comparing cached records to len(id_list), return if they are equal, and otherwise
pass on id_list with cached IDs removed to bigtable get_bytes_multi. With dupes removed, id_list is empty,
and the resulting row_scan to bigtable is unconstrained, meaning it either won't finish, or will return irrelevant data.
Duplicates aren't visible in the result, so deduping early avoid this issue and any others that might result from repeated keys.

[codecov]

Codecov Report

Attention: Patch coverage is 83.33333% with 2 lines in your changes missing coverage. Please review.

⚠️ Parser warning

The parser emitted a warning. Please review your JUnit XML file:

Warning while parsing testcase attributes: Limit of string is 1000 chars, for name, we got 2083 at 1:157236 in /home/runner/work/sentry/sentry/.artifacts/pytest.junit.xml

Files with missing lines	Patch %	Lines
src/sentry/utils/kvstore/bigtable.py	33.33%	2 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master   #91976      +/-   ##
==========================================
+ Coverage   81.60%   87.63%   +6.02%     
==========================================
  Files       10355    10355              
  Lines      587356   587463     +107     
  Branches    22586    22586              
==========================================
+ Hits       479306   514810   +35504     
+ Misses     107623    72226   -35397     
  Partials      427      427              

[wedamija]

It might be a good idea to put a warning or something inside of _get_bytes_multi too if we pass an empty list of ids

[kcons]

@wedamija I think all the other many/multi implementations do a reasonable thing with an empty list.. thoughts on just returning here, maybe with a warning that you probably didn't mean to ask for nothing?

[wedamija]

@wedamija I think all the other many/multi implementations do a reasonable thing with an empty list.. thoughts on just returning here, maybe with a warning that you probably didn't mean to ask for nothing?

That seems reasonable to me!