Shared parseCommonInputs utility -- scoped proof-of-concept for #378

[aaravkharwar-hub]

Addresses #378 -- but scoped intentionally, and I'd like your input on direction before going further.

What this is

parse_inputs() is duplicated across all 26 algorithm files, and most of that duplication is identical: init/initimg, verbose, qualmeas, gpuids, groundtruth. This PR extracts that shared portion into MATLAB/Utilities/parseCommonInputs.m, and migrates two algorithms (CGLS and SIRT) as a proof of concept, rather than attempting all 26 at once.

Each migrated algorithm's own parse_inputs now just calls parseCommonInputs for the common options, then parses its own remaining algorithm-specific options (e.g. CGLS's restart; SIRT's lambda, lambda_red, nonneg, redundancy_weighting) from the leftover argument list, exactly as before.

Why only 2 algorithms, not all 26

This touches option-parsing for real reconstruction algorithms that people run on real scan data, so I didn't want to blind-rewrite all 26 in one pass without your sign-off on the approach. A few things came up even in just these two that I think are worth your eyes before scaling this up:

1. A bug found and fixed: ind=double.empty(0,1) (present in every algorithm's parse_inputs) relies on MATLAB's classdef static-method syntax and isn't portable (fails under GNU Octave, for instance). Replaced with ind=[] -- identical behavior, more portable. I did NOT change this in the other 24 files, only the 2 migrated here.
2. A design subtlety: init_multigrid is not a shared utility -- it's duplicated as a local (file-private) function inside several algorithms (SART, SIRT, OSEM, OS_SART, SART_TV), each only visible within its own file. A shared parser calling init_multigrid(...) directly would silently break any algorithm whose own copy isn't globally visible. parseCommonInputs takes an optional function handle for this (multigridFn) so each algorithm can pass its own local implementation; SIRT does this (@init_multigrid, created inside SIRT.m so it correctly captures SIRT's own local function). CGLS doesn't have its own init_multigrid at all today (calling 'Init','multigrid' on CGLS already errors with "undefined function" before this PR) -- that's a pre-existing gap, unrelated to this refactor, and I've left it alone.
3. A preserved (not fixed) quirk: if 'init','image' is passed without 'initimg', every existing algorithm silently leaves the result empty rather than erroring. I kept this exact behavior in parseCommonInputs rather than "fixing" it, since this PR is meant to be structural only. Flagging it here in case it's worth its own issue.

Testing

I don't have a MATLAB license in my environment, so I used GNU Octave to actually execute the code rather than just writing it and hoping:

• Unit-tested parseCommonInputs directly (defaults, every option, error paths, the preserved quirks above).
• Side-by-side equivalence tests: ran the original and migrated parse_inputs for both CGLS and SIRT against 12-14 cases each and diffed every output field -- all matched, including error identifiers.
• A targeted test confirming the multigridFn handle actually resolves to the caller's own local function and not some other init_multigrid, using a deliberately "wrong" decoy to make sure.

I have not been able to verify against real MATLAB specifically, only Octave -- flagging that as a limitation.

If this approach looks right to you, I'm happy to continue migrating the remaining 24 algorithms in follow-up PRs. If you'd rather see a different shape (e.g. a struct-based options object instead of a leftover-args split, or algorithm name dispatch as you originally suggested), let me know before I put more algorithms through this pattern.