feat: memory profiling

[P403n1x87]

We add memory profiling support in the form of Python memory domain allocator tracking. Because this is performed via tracing, the new memory mode adds considerable overhead, as expected.

The new feature is implemented by attaching the frame stack to an allocation event, which is then retrieved when a corresponding free event on the same memory address is performed. This allows for an accurate detection of potential memory leaks. Data is emitted at the sampling rate, but only if recent allocations have caused a change in the maximum resident memory. This is to avoid reporting allocation data in cases when this does not correspond to an actual increase in memory usage.

Preliminary comparisons to similar tools show that Echion's memory mode introduces much less overhead, allowing an allocation-intensive program to run much more quickly (but still about 4 times slower).