Add full unit testing suite and CI ecosystem support

[grey-nearing]

Summary of Changes

1. Comprehensive Unit Test Suite (204 Tests, 0 Failures)

Adds dedicated unit test coverage across previously untested core modules:

• evaluation/metrics.py (test/test_metrics.py): Unit tests for all 14 hydrological metrics (NSE, MSE, RMSE, KGE, Alpha-NSE, Beta-NSE, Beta-KGE, Pearson-r, FHV, FMS, FLV, Peak-Timing, Missed-Peaks, Peak-MAPE), edge cases (all-NaN, empty slices, shape mismatches), and metric dispatchers.
• training/loss.py & regularization.py (test/test_losses.py): Tests for MaskedMSELoss, MaskedRMSELoss, MaskedNSELoss, MaskedCMALLoss, multi-frequency masking, and multi-target loss weighting.
• utils/cmal_deterministic.py (test/test_cmal_math.py): Mathematical validation for CMAL PDF/CDF distributions, PPF inverse-CDF calculations, and vectorized quantile monotonicity.
• utils/samplingutils.py (test/test_samplingutils.py): Tests for point prediction sampling (sample_cmal, sample_cmal_deterministic, sample_mcd), zero-threshold normalization, and negative value handling modes.
• modelzoo/ (test/test_models.py): Tests for FC MLP layers, PositionalEncoding, Regression / CMAL prediction heads, and BaseModel lifecycle.
• training/ (test/test_trainer.py, test/test_logger.py): Tests for optimizers, loss/regularization initializers, Logger training/validation lifecycles, and artifact image generation.
• datasetzoo/caravan.py & registry (test/test_caravan.py, test/test_datasetregistry.py, test/test_mfdata_loader.py): Tests for static attribute loading, CSV timeseries parsing, dataset registry dynamic dispatch, and CLI data loader.
• utils/ (test/test_logging_utils.py, test/test_lstm_utils.py, test/test_memory.py, test/test_plots.py): Tests for memory management, LSTM weight initialization, log filters, and hydrograph plotting.
• CLI Dispatchers (test/test_cli.py): Argument parsing and mode dispatching for run.py and run_scheduler.py.

2. Bug Fixes & Code Improvements

• googlehydrology/utils/samplingutils.py: Fixed UnboundLocalError during error message formatting when embedding dropouts are uninitialized.
• googlehydrology/datasetzoo/caravan.py: Fixed NotImplementedError with Dask .chunk('auto') when attribute tables contain string/object columns by specifying chunk({'basin': -1}).
• googlehydrology/training/__init__.py: Cleaned up obsolete reference to non-existent TiedFrequencyMSERegularization.
• test/test_multimet.py: Corrected mock assertion sequence to reflect scaler save-before-scale ordering.
• test/test_uncertainty.py: Guarded negative sample clipping check for deterministic CMAL quantile outputs.

3. Google Python Style Guide & Readability Compliance

• Line Length (§3.2): 100% compliant with 80-character maximum line length across all test suites, docstrings, comments, and fixtures.
• Import Organization (§3.13): Standardized three-block import grouping (Standard Library, Third-Party, First-Party) separated by single blank lines.
• Unused Imports Cleaned: Removed obsolete imports across all test files.
• Test State Isolation: Restored root logger handlers in test_logging_utils.py and isolated global FLAGS using absl.testing.flagsaver in test_mfdata_loader.py.

4. CI Ecosystem Modernization & Security Hardening

• Environment: Added environments/environment_cpu.yml for fast, lightweight CI test execution on Python 3.12.
• Branch Triggers: Added main branch to CI pull_request and push triggers in .github/workflows/pytest-ci.yml.
• Security Policy Compliance (zizmor):
  • Added top-level permissions: read-all.
  • Pinned GitHub Actions to immutable full commit SHAs (actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4.2.2, conda-incubator/setup-miniconda@d2e6a045a86077fb6cad6f5adf368e9076ddaa8d # v3.1.0).
• Configuration: Configured pyproject.toml with test markers (unit, integration, slow, gpu) and warning filters.

---

Verification

• Test Suite: All 204 unit tests pass in ~2 minutes on CPU.
• Statement Coverage: Core modules (googlehydrology.evaluation, googlehydrology.training, googlehydrology.utils, googlehydrology.modelzoo) >95% covered.