Use only SSE2 on prefix sum

This MR makes the manually vectorised implementation of the prefix sum use only SSE2 (max) ops, which are available in any x86_64 architecture.

Potentially solves MooreAnalysis#30 (closed)

mentioned in issue MooreAnalysis#30 (closed)

changed the description

added RTA label

added hlt1-throughput-decreased label

removed hlt1-throughput-decreased label

assigned to @peilian

This should be ready for a test.

/ci-test

added ci-test-triggered label

• [2022-03-04 21:45] Validation started with lhcb-master-mr#3874

resolved all threads

assigned to @rmatev

mentioned in issue Moore#397 (closed)

@dcampora what about the failing pipeline?

added 80 commits

• 4ee9fbea...0c4581f9 - 79 commits from branch master
• 7388371f - Use only SSE2 on prefix sum.

Compare with previous version

added hlt1-throughput-decreased label

removed hlt1-throughput-decreased label

unassigned @peilian

resolved all threads

assigned to @raaij and unassigned @rmatev