API call always retrieves *all* workflow runs

[isaac-canvas]

I've been trying this action out and we have a lot of workflow runs. It takes about an hour for the action to execute. This time is all spent in the request to the Github API because it there is no filtering on the workflow runs it retrieves.

This PR solve this problem by limiting the data retrieved by the API request to the date ranges provided. This will speed up the API request as well as the processing time when running the action. It also saves bandwith and processing time on GitHub's API server :D.

Locally this makes the action only take ~8 seconds when querying a week's worth of workflow runs, as opposed to ~1h.

Note: I think we could optionally add -F per_page=100 to the gh api command to always get the maximum results and make fewer requests. I didn't see any noticeable changes when adding this so I didn't include it.

[KittyChiu]

@isaac-canvas ❤ Thanks much for the contributions. Apology for late response. I had a look at the schema and created_at is a good choice for performance.
I originally decided with run_started_at instead of create_at was because the job can queue but not started.

I have two minds about the PR. If I merge it, then it will a break change from data consistency perspective. Your thought?