Refactor computeLRWorkDim (#45812)
Summary:
Move duplicated code for computing LRWork array dimention form CPU/CUDA implementation of apply_svd into LinearAlgebraUtils
Reduce common multiplication factor from 7 to 5, which according to the documentation should be sufficient for LAPACK-3.6+
From https://github.com/Reference-LAPACK/lapack/blob/122506cd8b6ce050a200920c3d4c0b153b150fd8/SRC/cgesdd.f#L186
```
RWORK is REAL array, dimension (MAX(1,LRWORK))
Let mx = max(M,N) and mn = min(M,N).
If JOBZ = 'N', LRWORK >= 5*mn (LAPACK <= 3.6 needs 7*mn);
else if mx >> mn, LRWORK >= 5*mn*mn + 5*mn;
else LRWORK >= max( 5*mn*mn + 5*mn,
2*mx*mn + 2*mn*mn + mn ).
```
Pull Request resolved: https://github.com/pytorch/pytorch/pull/45812
Reviewed By: walterddr
Differential Revision: D24100836
Pulled By: malfet
fbshipit-source-id: 0ca86aed25077c91cf60086ed301298381d5f628
