Add matrix_zero_tol

[ZedongPeng]

Fixes #72

Description

Introduces a new parameter matrix_zero_tol (default 1e-9) to control near-zero entry pruning in the constraint matrix.

Log Example:

---------------------------------------------------------------------------------------
                                    cuPDLPx v0.2.5                                    
                        A GPU-Accelerated First-Order LP Solver                        
               (c) Haihao Lu, Massachusetts Institute of Technology, 2025              
---------------------------------------------------------------------------------------
Dropped 1 near-zero entry (|value| <= 1.0e-09) from the constraint matrix
Problem: 27 rows, 32 columns, 82 nonzeros
Settings:
  iter_limit         : 2147483647
  time_limit         : 3600.00 sec
  eps_opt            : 1.0e-04
  eps_feas           : 1.0e-04

Running presolver (PSLP v0.0.4)...
  status          : REDUCED
  presolve time   : 0.000287 sec
  reduced problem : 21 rows, 28 columns, 71 nonzeros

Preconditioning
  Ruiz scaling (10 iterations)
  Pock-Chambolle scaling (alpha=1.0000)
  Bound-objective scaling
---------------------------------------------------------------------------------------
   runtime     |     objective      |   absolute residuals    |   relative residuals    
  iter   time  |  pr obj    du obj  |  pr res  du res   gap   |  pr res  du res   gap   
---------------------------------------------------------------------------------------
     0 0.0e+00 |  0.0e+00   0.0e+00 | 0.0e+00 0.0e+00 0.0e+00 | 0.0e+00 0.0e+00 0.0e+00 
   200 0.0e+00 | -4.7e+02  -3.6e+01 | 1.6e+00 5.5e-01 4.3e+02 | 5.1e-03 2.2e-02 8.5e-01 
   400 8.9e-03 | -4.6e+02  -4.6e+02 | 1.5e-01 6.7e-03 5.2e+00 | 4.6e-04 2.6e-04 5.6e-03 
   600 1.2e-02 | -4.6e+02  -4.6e+02 | 3.0e-03 6.2e-05 3.0e-02 | 9.7e-06 2.4e-06 3.2e-05 
---------------------------------------------------------------------------------------
Solution Summary
  Status                 : OPTIMAL
  Presolve time          : 0.000287 sec
  Precondition time      : 0.002035 sec
  Solve time             : 0.0156 sec
  Iterations             : 600
  Primal objective       : -464.7655834
  Dual objective         : -464.7353711
  Objective gap          : 3.247e-05
  Primal infeas          : 9.665e-06
  Dual infeas            : 2.445e-06

[ZedongPeng]

Hi @LucasBoTang . Should we remove the zero_tolerance in matrix_desc_t?

cuPDLPx/include/cupdlpx_types.h

Line 160 in 38432d5

double zero_tolerance;

[LucasBoTang]

Hi @ZedongPeng, Deleting zero_tolerance in matrix_desc_t is OK with me and makes code cleaner. The only risk I see is nnz getting too large and causing memory issues, but this trade-off is acceptable.