Remove intermediary JSON conversion from generateFromCpuSamples

[elliette]

Work towards #7917

Refactors generateFromCpuSamples to remove the intermediary JSON conversion step.

Before this PR:

generateFromCpuSamples went from vm_service.CpuSamples -> serialized to JSON -> deserialized to CpuProfileData:

• generates a JSON representation of the stack frames (and as a side effect the timeline tree)
  • processes each node in the timeline tree, creates a JSON representation of the stack frame and assigns the frame ID to the node
  • keeps track of any frames missing package URIs
• creates a JSON representation of the trace events
• creates a JSON representation for the trace object, which includes the stack frames and trace events
• does a bulk request for all the missing package URIs and updates the trace object with those URIs
• deserializes the JSON trace object to a CpuProfileData object
  • As part of this, determines the median sample period
  • creates a CpuStackFrame object from the JSON stack frame representation

With this PR:

generateFromCpuSamples goes from vm_service.CpuSamples -> CpuProfileData

• creates the profile metadata object
  • as part of this, determines the median sample period
• generates the stack frames (and as a side effect the timeline tree)
  • processes each node in the timeline tree, creates a CpuStackFrame for it and assigns the frame ID to the node
  • keeps track of any frames missing package URIs, then at the end does a bulk request for all the missing package URIs and updates the frames with those URIs
• converts the CPU samples to sample events
  • trace events are included in the sample events
• finally, constructs the CpuProfileData object

Note: I filed #9353 to refactor this method more to avoid side-effects - the call to generate the stack frames creates the timeline tree and assigns frame IDs to every node in the tree. This was true before this change as well. However, I didn't want to change too much at once in this PR.

[elliette]

Note: Added a benchmark test to try to measure performance gains. Unfortunately seems pretty minimal, but I do think the code is cleaner without this step.

Benchmarks before this change:

29814.323529411766
31181.30769230769
29489.323529411766
32825.639344262294
37737.24528301887

Benchmarks after this change:

29160.1884057971
29295.420289855072
29321.144927536232
29251.405797101448
29419.176470588234

[bkonyi]

Nice! Thanks for fixing my questionable code 😉