Step recorder with reversible trees

[WhiteBlackGoose]

So... a basic mechanism that is able to

1. Record steps while solving something
2. Reverse to the branch's root whenever we fail to solve something within a method

[Happypig375]

This is step-by-step solving 🦆

[WhiteBlackGoose]

Exactly 🦆

[Happypig375]

#28 🦆

[Rafael-SOWNet]

Part 1 of the two is implemented; part 2 is not. Measured on master at a45a7256.

1. Record steps while solving something

Done, by #819 and #951:

using var recording = RewriteRecording.Start();
var simplified = "a / (b / c)".ToEntity().Simplify();
// 259 steps, of which 13 name the individual rewrite that fired
// e.g. Common/Divf(var any1, Divf(var any2, var any3)): 1 / (1 / c) -> 1 * c / 1

A step carries the subexpression before and after, the rule set that rewrote it, the individual rule where the set is addressable, and the relation and soundness tier that set declares.

2. Reverse to the branch's root whenever we fail to solve something within a method

Not done. There is no API for reversing, and the recording is a flat sequence rather than a tree — it records what happened, not the branch structure that would let a caller unwind to a decision point and take the other route.

Worth noting what the recording did turn up about part 1's usefulness, since it bears on whether part 2 is worth building: of those 259 steps, 246 come from four sets that cannot name a rule at all, and CanonicalOrder alone is 91% of them. So a recording today is mostly normalisation noise. I raised that on #28 — a derivation that is 91% sort steps is not one a human reads, whatever the reversal mechanism looks like.

Suggest either re-scoping this to part 2 alone, or closing it in favour of #28, which is where the recording's grain is already being discussed.