[FIX] ASR: correct reconstruction dim count, blend seam, and RMS window rounding

[sappelhoff]

Three independent fixes in the calibration/reconstruction path:

• maxdims / keep-mask off-by-one in asr_process: maxdims used np.fix (truncate toward zero) instead of round-half-away, and the keep-mask index term protected one component too many. The spec converts the fractional 0.66 factor with round-half-away-from-zero, and its 1-based (1:C) < (C - maxdims) protects C-maxdims-1 components, so use int(np.floor(0.66 * nc + 0.5)) and np.arange(nc) < nc - maxdims - 1. Together these made the code under-remove artifact dimensions.
• Raised-cosine blend endpoint in asr_process: np.arange(ns)/ns never reaches 1, so the block's final sample kept a residual fraction of the previous reconstruction. Shift the phase to np.arange(1, ns + 1)/ns so blend[-1] = 1.0 and the last sample equals the new reconstruction.
• RMS window starts in asr_calibrate: the step N*(1-win_overlap) was rounded before building the window starts, drifting from the correct positions for a non-integer step (biasing mu/sig/T). Extract a _rms_window_offsets helper that rounds each start (matching clean_windows).

Testing

Adds a regression test for each. Full tests/test_asr.py passes; ruff clean.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 83.16%. Comparing base (3b0c530) to head (873385b).

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #111      +/-   ##
==========================================
+ Coverage   83.15%   83.16%   +0.01%     
==========================================
  Files          25       25              
  Lines        2832     2834       +2     
==========================================
+ Hits         2355     2357       +2     
  Misses        477      477              

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[sappelhoff]

---

Note: includes the nonlinear_eigenspace reproducibility fix from #118 (needed for the riemann path to have stable CI). Once #118 merges, this commit will drop out on rebase.