MILC interface: Exact current

[leonhostetler]

This is the second and final PR in the series to enable exact current calculations for HVP in the MILC/QUDA interface. The previous PR was #1590 .

Changes:

1. Adds a new MILC interface function, qudaExactCurrent, in lib/milc_interface.cpp. This computes the exact low-mode contribution to the current densities needed for HVP current calculations.

2. Implements eigenvalue shifting in the MILC interface for preserved deflation spaces. Previously, when using a preserved deflation space across multiple solves, the eigenvalues were recomputed each time the mass changed. With this update, computeEvals is called only once, during the first solve. The resulting zero-mass eigenvalues are cached and shifted as needed for subsequent masses.

   This avoids many unnecessary mat-vecs and greatly reduces output volume: eigenvalues are now printed only once instead of once per solve, which previously could produce O(10^6) lines of eigenvalue output in a typical production run.

3. Makes lib/eig_block_trlm.cpp and lib/eig_trlm.cpp slightly more verbose by printing progress notifications every 10th or 100th Lanczos step. On production-size lattices, eigensolves can otherwise run for more than an hour without any visible progress, which can make a normally progressing job appear to be stalled.

Testing:

• Verified exact-current results against the corresponding MILC-side implementation.
• Verified that preserved deflation spaces reuse cached zero-mass eigenvalues and shift them correctly for subsequent masses.

[maddyscientist]

This looks good to me. I'll leave @weinbe2 to comment on the MILC interface specifics. I've left a couple of comments.

[maddyscientist]

Thanks for the tweaks @leonhostetler. I've left a couple of non-essential comments that might be nice to update if you make another push.

@weinbe2 any thoughts before we merge?