Give the two AdaptiveLinearRegressorUnit implementations distinct names

[cboulay]

Two distinct public Units currently share the name AdaptiveLinearRegressorUnit:

• ezmsg.learn.process.adaptive_linear_regressor.AdaptiveLinearRegressorUnit: adaptive/incremental fitting.
• ezmsg.learn.process.linear_regressor.AdaptiveLinearRegressorUnit: uses LinearRegressorSettings and LinearRegressorTransformer; its documented fitting behavior resets/refits the model from the supplied training data rather than performing incremental adaptation.

The collision makes imports and component selection confusing. It also surfaced while adding ezmsg.components entry points: the registrations need module-qualified names to distinguish the two Units, and their short names are identical in the dashboard palette.

Suggested direction: rename the Unit in linear_regressor.py to LinearRegressorUnit (or BatchLinearRegressorUnit if that communicates the fitting behavior more clearly), retaining AdaptiveLinearRegressorUnit for the incremental implementation.

Acceptance criteria:

• Choose distinct public names that communicate the fitting behavior.
• Update internal references, examples, documentation, tests, and component entry-point declarations.
• Decide on a compatibility/deprecation path for the old import and registration names so existing pipelines and saved graph documents have a migration path.