Advance every training clock the compiled function reads

[jessegrabowski]

A schedule places itself in time by reading a StepCounter, but nothing advanced that counter unless the training step came from compile_train. Build the updates yourself and compile with function and the clock sat at zero, so the schedule returned its initial rate on every call -- under a warmup that rate is zero, and the network never moved while reporting a flat loss.

function now threads the one-step advance for every clock the graph reads, alongside the RNG updates it already threads. A caller's own updates entry still wins, which is how a clock gets pinned, and a pinned clock is exempt from the check that every clock agrees on its step count -- it isn't counting these steps, so it has nothing to agree with.

Closes #79
Closes #106

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 97.47%. Comparing base (559cfc8) to head (db79d07).

Additional details and impacted files

@@           Coverage Diff           @@
##             main     #105   +/-   ##
=======================================
  Coverage   97.47%   97.47%           
=======================================
  Files          60       60           
  Lines        3011     3012    +1     
=======================================
+ Hits         2935     2936    +1     
  Misses         76       76           

☔ View full report in Codecov by Harness.
📢 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.