Let the boolean table solver be cancelled

[Rafael-SOWNet]

Towards #858, and it removes most of the reason that issue needed a decision.

SolveBooleanTable and BuildTruthTable never consulted the cancellation token, so the escape hatch that works everywhere else in the library did not exist here. Measured before changing anything: a token cancelled after three seconds was still being ignored twenty seconds later.

That matters because of where the cost of these methods now sits. #864 made the search cheap, which leaves writing every model down as the wall — and how tall it is depends on the answer, not the question:

tautology	rows	time	memory
18 variables	262 144	575 ms	124 MB
20 variables	1 048 576	1.7 s	544 MB
22 variables	4 194 304	8.0 s	2.4 GB

A caller cannot know in advance which of those they asked for, and a formula every assignment satisfies offers nothing to prune. Being able to stop is the whole of the remedy available without changing the signature — which is why this is worth doing regardless of how #858 is settled.

What changed

MultithreadingFunctional.ExitIfCancelled() — the same helper PolynomialGcd, Simplificator, Minimiser and ExponentialSolver already use — at three places:

• the branch nodes of the pruning search;
• the loop that writes out completions, which is where the gigabytes land;
• the truth-table loop, which is all 2ⁿ rows by definition and so has nothing but stopping to offer.

It costs nothing measurable. 22 variables went from 7979 ms to 7943 ms — inside the noise — because the check is one async-local read set against a per-row allocation.

On #858's remaining question

Worth recording what this leaves. The genuinely useful additions — asking whether a formula is satisfiable, asking for one model, capping the count — are all additive, so they do not need the 2.0 window and can be designed at leisure in 2.1. The only thing that would need the window is changing SolveBooleanTable's existing return type, and there is a decent argument against: a method named for returning the table of solutions returning exactly that is coherent, and the bounded and lazy needs are better served by new members than by mutating this one.

So #858 is no longer a deadline, which is the useful part.

Verification

• cancellation returns at the moment it is asked, rather than not at all
• 6087 C# tests pass, 0 failed — three new
• 130 F# wrapper tests pass

No behavioural change for anyone who does not set a token, and no signature change, so nothing for BREAKING-CHANGES.md.

🤖 Generated with Claude Code