Do not call the caller's typo a library defect

[Rafael-SOWNet]

Entity.Set.SpecialSet.Create(string) and Entity.Set.SpecialSet.Create(Domain) are both public,
and both answered an input they did not recognise with AngouriBugException — whose message ends
"please report about it to the official repository". Somebody who writes "NN" is told their own
typo is a defect in this library and asked to file it.

UnrecognizedDomainException has existed for exactly this case since it was written, and nothing
threw it.

The second overload is the sharper one: Domain.Any is a documented member of the public enum.
It means no restriction, which is not a set this library has a node for — Domains.IsWithinDomain
answers it before ever reaching Create — so the only way to hit that arm is from outside, with
valid public API, and the answer was "report this as a bug".

Measured, on a build of each side

                        2.3.0                              this branch
Create("NN")            AngouriBugException  ...please  →  UnrecognizedDomainException:
                        report about it to the official     Unrecognized domain NN
                        repository
Create(Domain.Any)      the same                        →  NotSufficientlySupportedException:
                                                           There is no special set for domain Any
Create((Domain)99)      the same                        →  NotSufficientlySupportedException:
                                                           There is no special set for domain 99
Create("RR")            RR                              →  RR

NotSufficientlySupportedException rather than a parse error for the second, because the caller's
value is well formed and this library has no set for it — which is what that type means per
Docs/Usage/Exceptions.md.

Both stay under AngouriMathBaseException, so the catch the documentation tells a caller to
write is unaffected, and a test pins that. Nothing inside the library reaches either arm: the only
other route into the string overload is a SPECIALSET token, every spelling of which the switch
lists.

Verification

• 22 new tests in Sources/Tests/UnitTests/Core/UnknownDomainIsNotABugTest.cs, covering all ten
  accepted names, all five domains that are sets, both refusal paths, and the base-exception
  guarantee.
• Full suite on this branch: 7555 passed / 0 failed / 14 skipped, against a master baseline of
  7533 / 0 / 14 — exactly the 22 added, nothing else moved.
• BREAKING-CHANGES.md entry with both values, measured rather than read off the diff.

Merge order

Adds an ## Unreleased section to BREAKING-CHANGES.md, so it costs one mechanical conflict round
against any other branch that adds one — #1025 does. It also appends a file_header_template scope
to Sources/.editorconfig, which #1016, #1017, #1025 and #1027 also append to; all the entries
survive that resolution.

Found while documenting the exception hierarchy for
#746 item 11 (#1027), which correctly left
it alone because changing which type a site throws is a breaking change and that PR is not.

Related: #1028 is the same wrong claim in a different place — SumAll() on an empty sequence.