A fold over an empty sequence answers instead of throwing (#1028)

[Rafael-SOWNet]

Closes #1028.

new Entity[0].SumAll();   // AngouriBugException: At least 1 child required
                          // "...please report about it to the official repository"

An empty list — which a caller's own Where can easily produce — asked them to file a bug report
against this repository for their own data.

	before	now
new Entity[0].SumAll()	AngouriBugException	0
new Entity[0].MultiplyAll()	AngouriBugException	1
Sumf.Sum(new Entity[0])	AngouriBugException	0
Mulf.Multiply(new Entity[0])	AngouriBugException	1
MathS.Vector()	IndexOutOfRangeException	InvalidMatrixOperationException
new Entity[0].ToVector()	IndexOutOfRangeException	InvalidMatrixOperationException

Why 0 and 1 rather than a better exception

A fold over a monoid has an identity. That is not a preference — it is what makes

xs.Concat(ys).SumAll() == xs.SumAll() + ys.SumAll()

hold for every pair, the empty one included. The test pins that law across every split point
of a 0-, 1- and 3-element sequence rather than pinning the two constants.

IndexOutOfRangeException is a separate problem: it is not under AngouriMathBaseException, so a
caller catching the hierarchy Docs/Usage/Exceptions.md documents did not catch it at all.

Two of the five entry points are not named in the issue

The defect is passing an unchecked caller collection into MultiHangBinary, whose >= 1
precondition is genuine. Grepping its callers turned up Sumf.Sum(IReadOnlyList<Entity>) and
Mulf.Multiply(IReadOnlyList<Entity>) doing exactly that — both public, neither mentioned in the
report.

MultiHangBinary itself is unchanged. The issue is right that the bug is at the boundary rather
than in the fold, and its internal callers do rely on the precondition.

Fails 7 of 7 without the change. Full suite: 8758 passed, 0 failed.

🤖 Generated with Claude Code

https://claude.ai/code/session_01Bjumi5K7fg8yx6UK1mZTQd