Exercise 1: Only margin-left Is Auto — Possible Solution
====================================================================

THE TEST
------------------------------
    e = block({'width': '100px', 'margin': '0 20px 0 auto'})
    calculate_block_width(e, root_cb)   # root_cb content width = 300
    e.dimensions.margin.left
    e.dimensions.margin.right

RESULT
------------------------------
    margin.right -> 20.0   (unchanged, exactly as declared)
    margin.left  -> 180.0

WORKING THROUGH THE 4-VALUE SHORTHAND FIRST
------------------------------
'0 20px 0 auto' is a 4-value margin shorthand: top=0, right=20px,
bottom=0, left=auto -- read directly in that order (top, right,
bottom, left), no reuse rule applies here since all four positions
are given explicitly. expand_margin_shorthand produces
MarginEdges(top=0.0, right=20.0, bottom=0.0, left='auto').

WHY margin-left ALONE ABSORBS THE FULL LEFTOVER
------------------------------
calculate_block_width detects ml_auto=True, mr_auto=False. Since
width_str is '100px' (not 'auto'), the width_auto branch is skipped
entirely, and the "not width_auto" branch runs:

    if ml_auto and mr_auto:
        ...
    elif ml_auto:
        margin_left = underflow
    elif mr_auto:
        ...
    else:
        ...

Only the `elif ml_auto:` branch matches (mr_auto is False), so
margin_left is set to `underflow` directly -- the ENTIRE leftover
space, not half of it. underflow is computed as:

    total = margin_left(0, zeroed since auto) + border.left(0) +
            padding.left(0) + width(100) + padding.right(0) +
            border.right(0) + margin_right(20, NOT auto, kept as-is)
          = 0 + 0 + 0 + 100 + 0 + 0 + 20 = 120
    underflow = cb_width(300) - total(120) = 180

So margin_left becomes 180. margin_right, since mr_auto is False, is
never touched by any of the four branches at all -- it keeps its
originally parsed value of 20.0 straight from expand_margin_shorthand,
completely untouched by the auto-resolution logic.

WHY THIS WORKS AS AN ANSWER
------------------------------
This confirms the algorithm's own branch selection genuinely depends
on WHICH SPECIFIC combination of auto flags is set, not just "how many"
are auto. A single auto margin behaves completely differently from two
auto margins (which split the leftover evenly, per the chapter's own
Case C) -- with only one auto margin, that ONE margin takes on the
entire burden of making the equation balance, and the other, explicitly
-declared margin is left exactly as the author wrote it, never
recomputed or touched in any way.