reader/csv: avoid lseek for getting offset

[Riolku]

Currently we use lseek() after every row in the parallel CSV reader to get the file offset. I thought this would be very cheap, and figured it was easier than tracking it ourselves. However, it is not cheap at all, as demonstrated by a flamegraph, and it is not hard to track ourselves either.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Comparison is base (15f79e6) 92.90% compared to head (5d0ca82) 92.90%.

Additional details and impacted files

@@           Coverage Diff           @@
##           master    #2569   +/-   ##
=======================================
  Coverage   92.90%   92.90%           
=======================================
  Files        1026     1026           
  Lines       38591    38588    -3     
=======================================
- Hits        35853    35851    -2     
+ Misses       2738     2737    -1     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[Riolku]

Benchmark:

Query: LOAD FROM "100mil_pk.csv" RETURN COUNT(*);

The file contains 100 million integers, and nothing else.

master: 9.284s.
this branch: 2.427s.

[Riolku]

LDBC100 Comment:

master: 80.67s.
this branch: 57.85s.