dedupe code in all.py and single.py

[anthonyduong9]

Problem

• all.py and single.py both contain functions _get_safe_dtype() and _safe_cast()
• all.py and single.py duplicate get_layer_num_from_sae_id()
• A large block of code to calculate DFA values has been practically duplicated in all.py and single.py, and so it seems unlikely we'll want to change it in two places.

Fix

• Moving one pair of functions _get_safe_dtype() and _safe_cast() to shared.py and deleting the other pair.
• Moving get_layer_num_from_sae_id() to shared.py.
• Extracting the large block of code to calculate_per_source_dfa().

Testing

• I compared code I thought to be duplicated at www.diffchecker.com.
• I ran make check-ci.

[codecov]

Codecov Report

❌ Patch coverage is 80.00000% with 7 lines in your changes missing coverage. Please review.
✅ Project coverage is 8.49%. Comparing base (e8ddd08) to head (9c7f58b).

Files with missing lines	Patch %	Lines
apps/inference/neuronpedia_inference/shared.py	76.66%	5 Missing and 2 partials ⚠️

Additional details and impacted files

@@           Coverage Diff            @@
##            main     #76      +/-   ##
========================================
+ Coverage   7.86%   8.49%   +0.62%     
========================================
  Files        124     124              
  Lines      17470   17448      -22     
  Branches     381     379       -2     
========================================
+ Hits        1374    1482     +108     
+ Misses     16085   15950     -135     
- Partials      11      16       +5     

Flag	Coverage Δ
autointerp	100.00% <ø> (ø)
inference	45.02% <80.00%> (+8.24%)	⬆️
webapp	5.12% <ø> (ø)

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[hijohnnylin]

@anthonyduong9 thanks for the update, sorry for slowness on getting around to it!

we need test(s) that the before/after results are the same before merging this - can you write them, or let me know what info you'd need to write them? if you don't have time for it lmk - i can work on it.

[anthonyduong9]

@hijohnnylin I've added tests for calculate_dfa() and ActivationProcessor.process_activations(), since if you start from the functions where we extracted calculate_per_source_dfa() and go to where they're called, and so on, those're the public functions you end up at. They pass on main (save for one that references calculate_per_source_dfa(), since it doesn't exist), as well as this branch. The tests for calculate_dfa() largely match this in SAEDashboard, so assuming it's correct, we can have more confidence.