fix(tinytorch): tito crashes on Windows when a subprocess's output can't be decoded as cp1252

[Shashank-Tripathi-07]

What I did

Ran through the full TinyTorch student workflow on a real Windows machine (install -> setup -> modules 01-09 -> milestones 01-04 -> module test/reset -> community/dev commands) to hunt for the kind of fatal, no-recourse errors this project has had (see #1960, #1966/#1969).

The bug

tito module test 01 -- a command every student runs constantly -- crashes outright:

TypeError: sequence item 1: expected str instance, NoneType found

Root cause

tito/commands/module/test.py runs pytest via subprocess.run(..., capture_output=True, text=True) without an encoding=. On Windows, text mode decodes the subprocess's output using the console's legacy codepage (cp1252 here), not UTF-8. pytest's own output contains characters cp1252 can't represent, so the background thread that reads and decodes the pipe dies with an unhandled UnicodeDecodeError -- and subprocess.run silently gives back None for that stream instead of raising or returning a string. That None then blows up "\n".join(all_output), taking the whole command down with a raw traceback.

This is the same root cause already fixed in tito/main.py's own stdout/stderr (part of an earlier PR) and in install.sh, just in code neither of those touched. I checked the rest of the tito package for the same pattern and found it in 13 files, ~30 call sites -- every subprocess.run/Popen call using text=True except two in module/workflow.py that had already been fixed at some point. Since it's one root cause (no consistent subprocess text encoding policy), not 30 unrelated bugs, this fixes all of them in one pass: every text=True call now also specifies encoding="utf-8", errors="replace", so undecodable output gets replaced with a placeholder character instead of killing the reader thread.

Files touched: commands/dev/preflight.py, commands/dev/test.py, commands/export_utils.py, commands/milestone.py, commands/module/test.py, commands/module/workflow.py (the Jupyter-launch Popen that hadn't been covered), commands/nbgrader.py, commands/package/nbdev.py, commands/setup.py, commands/system/health.py, commands/system/update.py, core/auth.py, core/status_analyzer.py.

Testing

• Reproduced the crash on tito module test 01, confirmed the fix resolves it (runs clean, all phases pass).
• Re-ran the full workflow after the fix with no other crashes: tito setup, modules 01 through 09 (start/complete), milestones 01-04 (perceptron, XOR, MLP, CNN -- including real training), module test, module reset, module status, community status, dev export --help, and a sweep of every tito <group> --help plus deliberately invalid invocations (bad module numbers, missing arguments, unknown milestones).
• python -m py_compile on every changed file.