Address IT installation dependency conflicts (numpy/safetensors/torchmetrics/etc)

[speediedan]

Address installation dependency conflicts in IT where various packages (e.g., numpy, safetensors, torchmetrics) get installed, downgraded, and re-upgraded in different phases. For example:

• numpy 2.1.2 gets installed from FTS (and initial torch) install
• interpretune lightning install causes numpy 1.26 to be installed due to lightning/torchmetrics requirements
• manual install of numpy 2.3.0 (and safetensors) is needed to install circuit-tracer with no-deps
  This causes confusion and possible breakage. Unify/simplify the install process to avoid conflicting requirements and re-installations.

[speediedan]

resolved by #171