Fix LDSWORK being overwritten during workspace query in xTRSYL3

[nakatamaho]

Title: Fix LDSWORK overwrite during workspace query in xTRSYL3

## Summary

Remove the erroneous `LDSWORK = 2` assignment in the workspace query
branch of STRSYL3, DTRSYL3, CTRSYL3, and ZTRSYL3.

## Problem

When the caller passes `LDSWORK = -1` to perform a workspace query,
the routines overwrite LDSWORK with 2 before returning. This silently
corrupts the caller's variable, violating the documented intent(in)
contract. A subsequent call with the same variable would use
LDSWORK=2 instead of the caller's allocated leading dimension,
potentially triggering the unblocked fallback path or causing
out-of-bounds access.

## Fix

Delete the `LDSWORK = 2` line from all four routines. The assignment
was unnecessary: the only SWORK accesses in the LQUERY branch are
SWORK(1,1) and SWORK(2,1), both in column 1. In Fortran column-major
layout, the offset is `(i-1) + (j-1)*LDSWORK`; with j=1 the LDSWORK
term vanishes, so the stores are valid regardless of the LDSWORK value.

## Files changed

- SRC/strsyl3.f
- SRC/dtrsyl3.f
- SRC/ctrsyl3.f
- SRC/ztrsyl3.f

## Testing

No new tests required. Existing xTRSYL3 tests continue to pass.
The fix is a pure deletion of a dead store that only manifested as
caller-side corruption.

[langou]

I agree that

         LDSWORK = 2

does not work with the intent "input only" for LDSWOKRK. Thanks!

[langou]

The declaration of SWORK at

lapack/SRC/dtrsyl3.f

Line 211 in eb5f719

$ SWORK( LDSWORK, * )

with the variable LDSWORK set to -1 is interesting. I had not realized this was a thing.