Store the profile time range in the profile, rather than computing it based on samples / markers

[mstange]

When we load a profile, we compute the visible start and end time based on the sample and marker data in the profile.
It would be faster and less brittle to skip this computation.

I would say that, in all cases, the tool which creates the profile knows best for which time range it has data. So it should write that time range down in the profile.

This would also simplify our code, because we don't need to worry about caching the computed time range efficiently.
And it would fix bugs where the automatic computation gives bad results, such as #4129.
It would also make it easier to fix bugs in the uploading sanitizer, like #3518 and #3279.

┆Issue is synchronized with this Jira Task