All memory graphs should have the same scale

[julienw]

Currently memory graphs are computed independently. That means that in a situation like:

• memory graph A going from 0 to 1 GB
• memory graph B going from 0 to 10 kB

They would all look the same despite that memory graph B is unimportant.

We should compute one same scale for the whole profile and use it. I'm not sure yet whether this scale should change with the range, or rather stay fixed. Feedback welcome!

Then once we have this scale, we could also decide that some memory graphs could be automatically hidden depending on some threshold computed from the scale.

┆Issue is synchronized with this Jira Task

[gregtatum]

This means you need to visit every sample in the profile, which I think is why it's currently per-thread.