Release 2.0.0

[Rafael-SOWNet]

Names the version 2.0.0 instead of 2.0.0-preview.2, and drops "(unreleased)" from the
2.0.0 heading in BREAKING-CHANGES.md. That is the whole diff.

Why now

The two previews are published and the surface has settled. The last thing that moved
between them — the implicit conversion from Entity[] / List<Entity> / (Entity, Entity)
to Entity, which made every params Entity[] overload in the library uncallable with a
list — is removed and recorded, and nothing else in BREAKING-CHANGES.md is waiting on a
decision.

AssemblyVersion stays pinned at 2.0.0.0 and is deliberately not tracking Version: the
assembly is strong-named, so for a consumer without a binding redirect every
AssemblyVersion change breaks binding, and holding it makes each 2.x release a drop-in
replacement. The NuGet workflow takes the package version from the release tag, so Version
is what a local build and the assembly metadata carry.

Measured on this commit, .NET 10

C# tests	6253 passed, 0 failed, 14 skipped
F# tests	130 passed, 0 failed
casbench	117/119 with an elementary answer, 2 unsolved, 2 correctly refused, 0 wrong, 0 error, 0 timeout

What has to happen for the release itself

Merging this does not publish anything — Nuget.yml fires on a published release and takes
the version from the tag. So after merge:

gh release create v2.0.0 --title "v2.0.0" --latest --notes-file <notes>

That publishes AngouriMath, AngouriMath.FSharp, AngouriMath.Interactive and
AngouriMath.Terminal at 2.0.0 to nuget.org, and a pushed NuGet package cannot be
deleted, only unlisted — so it is deliberately left as one explicit step rather than done
from a branch.

The docs side, which is the other half of this

Two things go with the release and are not in this repository:

• The wiki is corrected and now machine-checked against a 2.0.0 build: 86 samples, 0
  compile errors, 0 output mismatches, 57 stated outputs verified. Before that, 23 of 90
  samples did not compile — Latexise had become Latexize, Exceptions documented the
  FutureReleaseException that Remove FutureReleaseException, which turned known gaps into bug reports #872 removed, and the entire F# page named a
  `dy/dx` that has always been `d/dx`.
• AngouriMathSite has a PR that
  drops --prerelease from the quickstart, removes the dead MyGet feed, adds the 2.0.0
  entry to What's new — which is where PackageReleaseNotes sends readers — and repairs
  the docs build, which has been broken since the library moved up a folder in f0db3eef
  on 2026-01-03: amsite.fsx and NaiveStaticGenerator both still published and read from
  Sources/AngouriMath/AngouriMath/. That PR should merge after the release, since its
  quickstart tells the reader to install a version that has to exist first.

Found while doing it, filed rather than smuggled in

• abs does not fold a negated radical or constant: abs(-sqrt(6)) survives Simplify #881 — abs folds a numeric argument and nothing else, so abs(-sqrt(6)) survives
  Simplify. It shows up in the documented answer to a quadratic inequality, and it
  falsifies the claim that An inequality with a symbolic coefficient gets an interval whose endpoints are ordered for one sign only #757's closed form "collapses to exactly the old output when the
  roots are concrete".
• Simplify rewrites the root node only: the same subexpression is left alone inside a matrix, a power or a sum #882 — Simplify rewrites the root node only: -(5 - sqrt(-11)) becomes sqrt(-11) - 5
  on its own and is left alone inside a matrix, a power or a sum. Not the complexity metric
  — the better form scores 7 against 9 — and FiniteSet works only because
  Simplificator.Alternate special-cases it.
• Docs tracking: examples in the almanac #585 is rewritten around a measurement instead of two merged pull requests: 184 of 1831
  documented members carry an <example>, and AngouriMath.Extensions has 0 of 89.