Upgrade Prefect to 3.1.13

[dgarros]

This PR is trying to upgrade Prefect to 3.1.13 because we need it for the event management epic as part of 1.2

Prefect made a lot of progress around Typing in 3.1.x but there are still a few issues reported by Mypy.
I opened some issues on prefect github project to track them and I've disabled mypy for these issues for now

• Mypy is reporting incompatible type for @task with arguments PrefectHQ/prefect#16769
• Mypy is reporting missing overload for @task and @flow that are part of a class PrefectHQ/prefect#16770

[codspeed-hq]

CodSpeed Performance Report

Merging #5518 will not alter performance

_{Comparing dga-20250119-prefect-3.1.13 (527d418) with develop (3cb8e6d)}

Summary

✅ 10 untouched benchmarks

[fatih-acar]

Need to investigate slowness on Prefect startup, it actually takes more than 30 seconds to start Prefect when running with xdist (probably because the CI runner is overwhelmed)

Maybe related to PrefectHQ/prefect#16299

[tomukmatthews]

Need to investigate slowness on Prefect startup, it actually takes more than 30 seconds to start Prefect when running with xdist (probably because the CI runner is overwhelmed)

Maybe related to PrefectHQ/prefect#16299

I have the same problem. I would be cautious about the latest versions of Prefect, many of our data pipelines are experiencing significant flakiness and degredation at the moment due to the performance related issues of the event driven architecture (internally) of prefect since 3.x. Have you experienced anything like that outside of tests?

[fatih-acar]

Need to investigate slowness on Prefect startup, it actually takes more than 30 seconds to start Prefect when running with xdist (probably because the CI runner is overwhelmed)
Maybe related to PrefectHQ/prefect#16299

I have the same problem. I would be cautious about the latest versions of Prefect, many of our data pipelines are experiencing significant flakiness and degredation at the moment due to the performance related issues of the event driven architecture (internally) of prefect since 3.x. Have you experienced anything like that outside of tests?

We started using Prefect since 3.0, so we can't tell for previous releases.
But there seems to be a small performance regression between 3.0 and 3.1 that makes our CI pipelines fail due to load on our runners that causes slow startup.
We observe startups taking between 20s to 30s, and sometimes reaching the default timeout value which is 30s.
Fixed that by increasing server_startup_timeout in prefect_test_harness to 60s.