viable/strict/1775563420: [memory viz] Restructure Memviz and add some tests (#179488)

The PyTorch development team has merged a significant internal refactor of their MemoryViz visualization tool, aimed squarely at improving code quality and testability. The core change in pull request #179488 was architectural: data-processing functions were extracted from the main `MemoryViz.js` file and moved into a new, standalone module called `process_alloc_data.js`. This strategic separation decouples the business logic of processing memory allocation data from the presentation layer that depends on d3.js and the browser's DOM. The move enables developers to write and run unit tests for the data-processing logic in isolation, a critical step for ensuring future changes don't introduce regressions.

The PR, authored with assistance from Claude AI, introduced a suite of new unit tests that validate key functionalities like allocation/free operation matching, the behavior of private memory pools, and the generation of segment snapshots. A new CI workflow was also added to run these tests automatically. Crucially, the team confirmed that this was purely a backend refactoring—the visual output of the MemoryViz tool remains identical before and after the changes, as shown in the provided screenshots. This work lays a foundational layer of testing for a tool that helps developers debug and optimize memory usage in PyTorch models, a constant challenge in large-scale AI training.