Improve API consumption of _handle_bulk_insert_op

[mr0re1]

• Instead of doing globalOperations.aggregateList perform series (usually just 1) of zonal list requests - to stop hitting HeavyWeightReadRequestsPerMinutePerProject limits;
• Do not perform server-side filtering on non-indexed fields, instead do it client-side to avoid ListRequestsFilterCostOverheadPerMinutePerProjectPerRegion limits;
• Incorporate createTime range and user (both are indexed) into server-side filtering.

New approach consumes 0 - filtered_list_cost_overhead, while old one was effectively churring (on server side) through all operations.

Latency is smaller but not significantly (~2 times improvement) - no thorough sampling has been done.

See: https://cloud.google.com/compute/api-quota#api-rate-limits

[tpdownes]

I'm concerned we are missing something in the case when multiple allowed zones are set by the user.