DLR creation memory usage is linear in $\beta$ (should be $\log\beta$)

### Description

This is a reposting of the triqs issue: TRIQS/triqs#917 showing that the memory usage building the DLR basis is linear in the inverse temperature $\beta$, preventing the use of DLR at low temperatures. This is probably caused by the current approach to select DLR Matsubara frequencies, since a dense Matsubara grid is used with an upper cutoff proportional to $\beta$.

See the memory scaling test here: TRIQS/triqs#917

**Expected behavior:** The peak memory usage should be $\log \beta$ for the DLR basis to be applicable in the low temperature regime.

**Actual behavior:** Peak memory usage is linear in $\beta$