fix output

[mohanchen]

Summary

Cleans up several runtime log messages and fixes a latent bug in the GPU memory tracker.

Changes

Bug fix

• Memory::print GPU path: previously record_gpu called print(find), but the body of print(int) indexed the CPU arrays (name[find], consume[find]). GPU memory warnings therefore printed the wrong name/size. Refactored print to take (name, size_mb) so the caller passes the correct array, fixing CPU and GPU paths uniformly.

Behavioral change

• Raised the memory-allocation warning threshold from 5 MB → 20 MB to reduce log noise. Allocations in the 5–20 MB range no longer emit the warning.

Output polish

• Reworded the memory warning header to *** Memory Allocation Warning ***.
• SET NONLOCAL PSEUDOPOTENTIAL PROJECTORS now appends the element label, so the line is meaningful in multi-element runs.
• START CHARGE value (init_chg) is now uppercased for consistency with surrounding output.
• For nspin == 2, the printed k-point count now shows the physical number (nkstot / 2) instead of the spin-duplicated total stored internally. nspin == 1 and nspin == 4 are unaffected, matching how set_kup_and_kdw populates nkstot.
• Added a blank line before the per-element pseudopotential block in read_cell_pseudopots for readability.
• Switched the "readin total magnetization" line in CalAtomsInfo to use ModuleBase::GlobalFunc::OUT so its formatting matches neighboring entries.