Changes proposed in this PR:

• Add climada.util.interpolation.sparse_local_exceedance and sparse_local_frequency, which compute the exceedance curve of every matrix column from that column's nonzero entries, instead of calling preprocess_and_interpolate_ev once per column on a densified CSR column.
• Call them from Impact.local_exceedance_impact, Impact.local_return_period, Hazard.local_exceedance_intensity and Hazard.local_return_period. supports_sparse_fast_path gates the switch; every other configuration keeps the existing code path unchanged.

getcol on a CSR matrix scans the whole index array, so calling it once per column made these methods O(n_columns × nnz).

Measured on a real US tropical cyclone dataset (LitPop USA 150 arcsec x historical TC hazard USA, both from the data API; 3,890 events x 643,099 exposure points) with pr_benchmark_exceedance.py, which times both paths in one process and compares them:

                                        old          new
Impact.local_exceedance_impact      62700 ms      869 ms    72x
Impact.local_return_period          63241 ms      882 ms    72x
Hazard.local_exceedance_intensity  109259 ms     1173 ms    93x
Hazard.local_return_period         109263 ms     1210 ms    90x

In a full risk assessment on that data the step goes from 83.6% of the run to 6.6%, and the run from 74.7 s to 13.1 s.

The script compares every pair with np.testing.assert_array_equal: no differences.

One thing worth flagging: the zeros can be dropped because they cannot influence a retained point. Sorted ascending, the zeros come first, and the cumulative frequency is a reverse cumulative sum, so the entry for the p-th largest value only ever sums the p largest frequencies. value_threshold=0 then discards exactly the entries the zeros sat in. This is why the helpers read data[start:end] rather than densifying; the reasoning depends on the threshold being 0, which is one of the conditions in supports_sparse_fast_path.

PR Author Checklist

• Read the Contribution Guide
• Correct target branch selected (if unsure, select develop)
• Descriptive pull request title added
• Source branch up-to-date with target branch
• Documentation updated
• Tests updated
• Tests passing
• No new linter issues
• Changelog updated

PR Reviewer Checklist

• Read the Contribution Guide
• CLIMADA Reviewer Checklist passed
• Tests passing
• No new linter issues