Restore per-cluster view spike budgets

[rossant]

Summary

• make Waveform, Amplitude, and Correlogram sampling fixed per cluster by default while retaining optional shared total caps
• prevent dataset-local view state, including the Firing Rate time range, from leaking through global GUI state
• clarify budget semantics, Firing Rate behavior, configuration examples, and troubleshooting documentation

Why

A shared total budget makes per-cluster sampling accuracy decrease as more clusters are selected. Both policies are useful, so the per-cluster ceilings remain primary and an integer *_total value opts into a shared cap. Separately, a stale globally persisted Firing Rate x_max could clip spikes in fresh datasets.

Validation

• make lint
• make format-check
• make doc-check
• core/GUI phase of make test-full: 363 passed, 1 skipped
• application suite against phylib master db10401589b47fd8d56f7c66531b0c22fc7f9e91: 133 passed
• focused GUI state and histogram regressions: 8 passed

The locked PyPI phylib 2.7.0 predates the existing spikes_are_disjoint selector argument already used on master; current phylib source is required for the application suite.