Heterogeneity index update (#62)

## Summary

Updates the heterogeneity index implementation to match the generalized
methodology.

- computes heterogeneity for the output and all inputs in one call
- normalizes regional sensitivity profiles before calculating H
- supports continuous and categorical custom partitions
- excludes a categorical input from its own regional sensitivity profile
- skips H_Y for categorical outputs with an explanatory warning
- exposes raw and normalized profiles, regional SI sums, region counts,
  and individual contributions through H.details
- adds plotting from the returned HeterogeneityResult object
- uses a minimum regional sample size of 100 observations
- adds unit tests for the new behavior

## API

Basic analysis:

```python
H = simdec.heterogeneity_indices(
    output=y,
    inputs=X
)
```

Custom partition:
```python
H = simdec.heterogeneity_indices(
    output=y,
    inputs=X,
    custom_partition=Z
)
```

Results:
```python
H.indices
H.details["X1"]
H.plot("X1")
```

Author: gnopik