Add Hermitian Support for Finite-Shot Measurement Statistics

[multiphaseCFD]

Context:

Description of the Change:

Benefits:

Possible Drawbacks:

Shortcut Stories:
[sc-53754]

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 97.95%. Comparing base (bd09fe1) to head (71af0cf).

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #451      +/-   ##
==========================================
- Coverage   99.95%   97.95%   -2.01%     
==========================================
  Files          20       71      +51     
  Lines        4439    10505    +6066     
  Branches        0      886     +886     
==========================================
+ Hits         4437    10290    +5853     
- Misses          2      170     +168     
- Partials        0       45      +45     

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

[dime10]

Thank you @multiphaseCFD 🥳

I think we also need to make sure that our wheels are built correctly. I would imagine that we need to add the lapack library to our binary distribution, unless you link it into lightning statically? Or does lightning look for it at runtime?

We also need to ensure the change works on MacOS (installing the liblapack-dev equivalent etc), running the wheel build action for mac from this branch would be a good test for it.

[dime10]

If we're adding the package here it probably needs to be added to all our other actions as well (where appropriate).
Our install docs also need to reflect this new dependency:

catalyst/doc/dev/installation.rst

Line 66 in 300ff89

Building from source

[dime10]

Can we target the minimum commit on the master branch that introduced support? Having a fixed hash will allow us to go back through Catalyst's git history and still build it successfully.

[dime10]

This option is added but unused?

[multiphaseCFD]

Thanks @dime10 ! Per packaging/wheels building, we should add RPATH (path\to\liblapack or scipy) to cmake. RPATH embeds information in the binary about where to find its dependent shared libraries at runtime. We've already use it to fix DSO errors for L-GPU when linking against custatevec.so.

[dime10]

Thanks @dime10 ! Per packaging/wheels building, we should add RPATH (path\to\liblapack or scipy) to cmake. RPATH embeds information in the binary about where to find its dependent shared libraries at runtime. We've already use it to fix DSO errors for L-GPU when linking against custatevec.so.

The one problem I see with this is that for deployed packages (like a Catalyst wheel), the correct RPATH (on the user's system) is not known at build time.