Solve size_batch loading time analytically when sizing by V_max

[sarangbhagwat]

Fixes #249.

Fix

The fixed-point equations have an exact analytical solution: the quadratic's discriminant collapses to a perfect square, giving

tau_loading = V_max * V_wf / F_vol

— the time to fill a reactor at maximum working volume under constant filling. This PR uses it directly:

if N_reactors is None:
    # Time required to load a reactor filled to maximum working volume
    # (analytical solution at a continuous number of reactors)
    tau_loading = V_max * V_wf / F_vol

    # Minimum number of reactors; ceil ensures N_reactors >= 2
    N_reactors = ceil(F_vol * (tau_reaction + tau_cleaning + tau_loading) / (V_max * V_wf))

The subsequent (already-existing) recompute of V_T, V_i, and tau_loading from the integer N_reactors is unchanged. Consequences:

• No iteration, no spurious root, no ZeroDivisionError.
• N_reactors = ceil(F_vol*(tau_reaction + tau_cleaning)/(V_max*V_wf) + 1) is the minimum count satisfying V_i <= V_max, and is always >= 2, so the max(N_reactors, 2) clamps are removed as redundant.
• The unused from flexsolve import wegstein import is dropped.
• Docstring notes updated; the stale constant-loading doctest updated to the actual (integer-N-consistent) output.

Verification

All on Python 3.14.7 / thermosteam 0.53.5:

1. Doctests (CI flags NORMALIZE_WHITESPACE IGNORE_EXCEPTION_DETAIL NUMBER ELLIPSIS):
   pytest --doctest-modules biosteam/units/design_tools/batch.py — passes (fails on master);
   pytest --doctest-modules biosteam/units/nrel_bioreactor.py — passes (the N=8 docstring example is byte-identical, confirming the N-given path is untouched).
2. Issue reproduction (NRELEthanolFermentation with tau=8, V_max=3785, the docstring feed):
   • 2.53.11: ZeroDivisionError in size_batch.
   • This branch: simulates cleanly — 2 reactors, 1727 m³ each (≤ 3785 m³ cap), loading time 11 hr, batch time 22 hr; self-consistent (working volume 1554 m³ = 141.3 m³/hr × 11 hr, and each reactor loads for exactly the 8 + 3 hr the other spends reacting + cleaning).
   • Identical numbers to current master's clamped-iteration result, so no behavior change where master already works.

[yoelcortes]

I double check the math by hand. I never realized it had a closed solution until now. Good catch!!