Improve jdbc performance

[loicmathieu]

Improve JDBC performance by lowering the min poll loop interval and changing PostgreSQL indexing.

Min poll loop interval is the minimum delay between each task execution. When running a single flow that have a lot of tasks, lowering it lower the delay between each tasks. It is currently configured at 100ms, configuring it to 25ms show a nice performance gain and should still be not too low so the database will not be sollicited too much. See performance gains bellow.

On PostgreSQL, the queues table unique index disturb the PostgreSQL cost optimizer that chooses this index when doing the poll query resulting in 10ms poll queries on a table with a few thousands of line.
Removing the PK and the index that exists on the columns type and offset, and replacing them with a hash index on the offset colm-umn do the trick. There is still an index on the offset column but it is no longuer used when doing the poll query. This new index is needed for other queries.

Using this flow;

id: slow-flow
namespace: io.kestra.tests

tasks:
  - id: waterconsumption
    type: io.kestra.core.tasks.flows.Template
    namespace: io.kestra.tests
    templateId: for-each-1
  - id: watertemperature
    type: io.kestra.core.tasks.flows.Template
    namespace: io.kestra.tests
    templateId: for-each-2
  - id: internaltemperature
    type: io.kestra.core.tasks.flows.Template
    namespace: io.kestra.tests
    templateId: for-each-3
  - id: internalhumidity
    type: io.kestra.core.tasks.flows.Template
    namespace: io.kestra.tests
    templateId: for-each-4

That using this template (all 4 templates are the same):

id: for-each-1
namespace: io.kestra.tests
tasks:
  - id: for-each-1
    type: io.kestra.core.tasks.flows.EachSequential
    tasks:
      - id: each-value-1
        type: io.kestra.core.tasks.debugs.Return
        format: "{{ task.id }} with current value '{{ taskrun.value }}'"
    value: "[\"value 1\", \"value 2\", \"value 3\", \"value 4\", \"value 5\", \"value 6\", \"value 7\", \"value 8\", \"value 9\", \"value 10\"]"

Gives the following results:

Baseline: 22s
Min poll interval from 100ms to 25ms: 15s
Changing PG indexing: 7s

These two optimizations on top of each other gives a 70% execution time reduction (execution time divided by 3 approx).

Even if not tested thoroughly on EachParallel, a quick test switching EachSequential to EachParalle show also a nice improvement.