Upgrade jaxdf to 0.3.0, fix FD accuracy bug (#224)

[astanziola]

Summary

• Bump jaxdf dependency from >=0.2.8 to >=0.3.0 (also pulls in jax 0.9.x, equinox 0.13.x)
• Regression test for simulate_wave_propagation solver error when using accuracy that's not the default accuracy of 8 #224: FiniteDifferences with accuracy != 8 and smooth_initial=False caused pytree mismatch in lax.scan. Verified: test fails with jaxdf 0.2.8, passes with 0.3.0.
• Fix CBS normalization putting JAX arrays into Domain static metadata (broke under JAX 0.9 stricter pytree checks)
• Relax BLI off-grid sensor test tolerance: float32 3D FFT interpolation precision varies across platforms (macOS Accelerate vs Linux MKL/OpenBLAS). Max deviation ~1e-5, was passing marginally with default rtol=1e-5, now uses rtol=1e-4.

Also included in jaxdf 0.3.0:

• Fixed inconsistent shift_operator between FourierSeries and FiniteDifferences (jaxdf#146), used in jwave for density staggering
• CUDA 12.8 compatibility via jax 0.9.x (Jwave is not compatible with cuda 12.8 #241)

Closes #224, closes #241.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 80.44%. Comparing base (bf8a0aa) to head (8201eab).
⚠️ Report is 1 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #246      +/-   ##
==========================================
+ Coverage   80.42%   80.44%   +0.01%     
==========================================
  Files          15       15              
  Lines        1216     1217       +1     
==========================================
+ Hits          978      979       +1     
  Misses        238      238              

☔ View full report in Codecov by Sentry.
📢 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.