reload functionality

[dvolgyes]

The regular tensorboard approach with event accumulators has a reload functionality.
That would be most helpful for dynamic situations where the data changes,
or new directories are created (see the children).

It doesn't have to be automatic, but it would be nice to make a .reload() which can update
things, instead of deleting the object and creating a new one.

[j3soon]

Are you encountering slow parsing speed when dealing with large event logs?

Exposing the EventAccumulator.reload function could save some time by allowing us to avoid loading from scratch. However, this would still require re-parsing everything in the EventAccumulator, and since the bottleneck currently lies in the parsing logic, I don't think this would significantly improve performance.

Despite this, I think it's worth adding this feature to help dealing with larger files. I may consider implementing it after addressing the current parsing bottleneck (#8).