Optimize memory usage in `homogeneous_DUO_update` via in-place update

[EwoutH]

This commit resolves a memory consumption issue in the homogeneous_DUO_update method by replacing the previous array reallocation with an in-place update using np.place. The old implementation created new NumPy arrays in each update cycle, leading to unnecessary memory overhead.

With the in-place update, total memory usage was reduced from 1189 MB to 471 MB.

Changes

• Switched to np.place for in-place updates of s.route_pref.
• Removed unnecessary temporary array creation to reduce memory footprint.

Part of https://github.com/toruseo/UXsim/issues/143.

[EwoutH]

@toruseo The failure values look very small, could it be that just some rounding error and the tolerances are a bit tight?

[toruseo]

The diff is not small. I guess (I dont fully understand the vectorizatoin technique used here) the modification broke the route choice logic.

[EwoutH]

Yes, took a look again, you’re right.

I will try a few other approaches, and also investigate dist_record. Could it be that s.route_pref is copied somewhere and/or pointed to from other entities?

[toruseo]

Closing, as better one is implemented by https://github.com/toruseo/UXsim/pull/146