Precise busy computation / stop counting when fiber is resolved

[TheLortex]

• Because the suspend event was not handled, the last fiber before suspension would be considered busy even if it was sleeping.
• Runtime events provide precise timers, we can use them to know when fiber switches happen, it will be more precise then getting the current timestamp when the event is read from the queue. This way, we also avoid using refresh_active_tasks which was in charge of incrementing the busy counter of the active fiber. Instead the total busy value is known when the fiber infos are rendered.
• When a fiber resolves, the idle value keeps increasing. Instead we can stop the counter when that happens.

I have also added text colors depending on the fiber state:

• white: active and paused
• bold: active and busy
• gray: resolved

[patricoferris]

This should fix https://github.com/patricoferris/meio/issues/27 ?

[TheLortex]

There's still a decision to take of when should they be collected, i.e. disappear from the screen ?

It's possible to implement a flexible strategy: never GC them, but hide them after a fixed delay, and add a way for the user to bring them back if needed.

[patricoferris]

It's possible to implement a flexible strategy: never GC them, but hide them after a fixed delay, and add a way for the user to bring them back if needed.

Whatever we decide don't let it block this PR :))

Hiding after a fixed delay seems reasonable, and until your extremely nice tree-view lands I don't think people will want to ever see them again as the console will be pretty hard to understand. We still might need some GC for Fiber.fork-heavy code if the table gets massive.