[BUG] .sif is accepted by io::read but only parses MPS, is untested, and is missing from the routing docs

[ramakrishnap-nv]

Describe the bug

io::read accepts .sif and routes it to read_mps():

if (lower.ends_with(".mps") || lower.ends_with(".qps") || lower.ends_with(".sif")) {
  ...
  return read_mps<i_t, f_t>(path, fixed_mps_format);
}

There is no SIF parser. SIF (the format CUTEst/SIFDEC problem sets ship in) uses VARIABLES, GROUPS and CONSTANTS sections where MPS uses ROWS, COLUMNS and RHS, so a genuine SIF file cannot be read by an MPS reader. The extension is effectively an undocumented alias for "parse this as MPS".

Three separate problems:

1. It only works for files that are already MPS. Verified against a 26.10 nightly libcuopt through cuOptReadProblem:

   file	content	result
   model.mps	MPS	CUOPT_SUCCESS, 2 vars / 1 constraint
   renamed.sif	identical bytes, .sif name	CUOPT_SUCCESS, 2 vars / 1 constraint
   real.sif	genuine SIF (VARIABLES/GROUPS/CONSTANTS)	CUOPT_MPS_PARSE_ERROR (3)

2. No test coverage anywhere. grep -rIn "\.sif\|SIF" cpp/tests python/cuopt/cuopt/tests java/cuopt/src/test returns nothing, and the repo contains no .sif fixture. Nothing would catch a regression, in any language.

3. parser.hpp contradicts itself. The Routing block in the doc comment lists only .mps/.qps/.lp, while the code matches .sif and the logic_error message enumerates .sif, .sif.gz, .sif.bz2, .sif.lz4.

Steps/Code to reproduce bug

cuOptOptimizationProblem p = NULL;
cuOptReadProblem("real.sif", &p);   // CUOPT_MPS_PARSE_ERROR, with a valid SIF file

Reproduces through every interface, since cuOptReadProblem (cuopt_c.cpp:292), the Python Read path (cython_parser.cpp:19) and the Java bindings all call the same io::read.

Expected behavior

Either support SIF properly — a parser, a fixture, and tests — or drop the extension. Advertising it in the error message while the routing docs omit it, with nothing tested behind it, is the worst of the three options.

Additional context

.sif entered in #1643 ("Include Q matrix in Ruiz scale computations", merged 2026-07-31), which is unrelated to file formats and does not mention .sif in its description. git log -S'.sif' -- cpp/ returns that commit alone, and no other issue or PR discusses SIF support, so this looks unintentional rather than a designed feature.

Noticed while documenting the read formats for the Java bindings (#1524), after @tmckayus asked what .sif is: #1524 (comment)

cc @rg20

[chris-maes]

We support .SIF to make it easy to read QPs form the Maros-Meszaros test sets. These .SIF files are effectively .QPS files. I don't think any change is necessary.