Allow converting an ALL_ALLOCATIONS capture file to AGGREGATED_ALLOCATIONS

Allow converting a capture file containing information about every allocation to one that contains only information about high water mark and leaked allocations. This could drastically shrink the file, at the cost of throwing away lots of (seldom used) information. (It's also quite useful for convincing yourself that the ALL_ALLOCATIONS format is correct 😅)

https://github.com/godlygeek/memray/tree/transform_to_aggregated contains a working implementation of this, but it's a tough sell architecturally, as it duplicates some code and couples the reader to the writer.

We'll need to decide whether we want this feature enough to justify cleaning it up and implementing it in an architecturally justifiable way.

Review the working implementation in the transform_to_aggregated branch, focusing on how the capture-file reader and writer interact. Decide on an architecturally justifiable approach that avoids duplicated code and tight reader–writer coupling; done means an ALL_ALLOCATIONS file can be converted to AGGREGATED_ALLOCATIONS while retaining only high-water-mark and leaked-allocation information.

Written by the indexing model from the issue text.