New-lambda computation `htpynewlam` ineficient

[mpf]

The profiler shows lots of time spent in htpynewlam, particularly in these lines:

https://github.com/MPF-Optimization-Laboratory/ActiveSetPursuit.jl/blob/cefb18f2eff5e0c1ee0e0fb7b25893d8a0b574a8/src/helpers.jl#L314-L326

These could be rolled into a single allocation-free loop.