Deprecate initialize_data_collector

[EwoutH]

Deprecate the Model's initialize_data_collector, which was added in #1287 to solve #1221.

Back when we had a schedulers it was logical, but now you can't forget to add Agents to a scheduler since that happens automatically.

The other reason was the batch_run assuming a Model attribute called datacollector to be present. An error in batch_run itself is far more robust to enforce that behavior, so that's added..

It also removes the behavior of directly collecting after init. Now you can do it when you want. It makes using the DataCollector explicit.

The migration guide was updated. Basically, replace:

self.initialize_data_collector(...)

With:

self.datacollector = DataCollector(...)

[github-actions]

Performance benchmarks:

Model	Size	Init time [95% CI]	Run time [95% CI]
BoltzmannWealth	small	🔴 +5.2% [+4.3%, +6.1%]	🔵 -0.1% [-0.3%, +0.0%]
BoltzmannWealth	large	🔵 +0.2% [-0.2%, +0.6%]	🔴 +4.6% [+3.4%, +6.0%]
Schelling	small	🔵 +0.3% [-0.1%, +0.7%]	🔵 +0.1% [-0.1%, +0.3%]
Schelling	large	🔵 +2.2% [+1.1%, +3.2%]	🔵 +2.2% [-0.4%, +4.7%]
WolfSheep	small	🔵 +1.5% [+1.3%, +1.7%]	🔵 +0.8% [+0.6%, +1.0%]
WolfSheep	large	🔵 +0.4% [-0.9%, +1.6%]	🔵 +0.3% [-4.2%, +4.6%]
BoidFlockers	small	🔵 +2.8% [+2.4%, +3.2%]	🔵 +1.6% [+1.0%, +2.3%]
BoidFlockers	large	🔵 +3.3% [+2.5%, +4.1%]	🔵 +2.2% [+1.6%, +2.7%]

[rht]

The other reason was the batch_run assuming a Model attribute called datacollector to be present. An error in batch_run itself is far more robust to enforce that behavior, so that's added..

Another reason is that there could be multiple data collector objects. As of today, I don't think there are any reasonable use cases that necessitate multiple data collectors (each with own scheduler/agent containers) anymore.

I'm fine with reducing the API surface, i.e. deleting this method.