Adopt ordered multi-caret markers in FCS tests (no behaviour change)

[T-Gro]

Follow-up to #20033, building on @auduchinok's #20077. Adopts the ordered multi-caret / selection markers (SourceContext.extractOrderedMarkedSources + {selstart}/{selend}) to remove copy-pasted test sources and magic coordinates.

No behaviour change — every migrated test keeps identical assertions (each commit rubber-duck-verified byte-for-byte; tests green on net472 + net10.0).

Reductions:

• Goto-def tests: drop unused open System + now-redundant caret-anchor (*loc-N*) comments
• Completion ArraySliceNotation (3 source copies → 1), CurriedArguments (5 → 1), Bug312557_2 (4 → 1), Bug69673_1 (2 → 1)
• Symbols Nested copy-and-update (8 Facts sharing a copied source → 1)
• BreakpointLocation: magic ((line,col),(line,col)) tuples → {selstart}/{selend} markers in the source

Deliberately not migrated (it would make the code worse, not better):

• Tooltip/Completion tests using markAtStartOfMarker/markAtEndOfMarker on a shared source — they already reuse the source, with heterogeneous asserts per position; {caretN} there is 0 reduction and messier.
• $ / direct-column low-level marker tests (IdentifierIsland, QuickParse, …) — $ is more compact than {caret}; migrating adds indirection + a ±1 column risk for no reduction.

[github-actions]

⚠️ Release notes required, but author opted out

Warning

Author opted out of release notes, check is disabled for this pull request.
cc @dotnet/fsharp-team-msft

[auduchinok]

Deliberately not migrated (it would make the code worse, not better):
Tooltip/Completion tests using markAtStartOfMarker/markAtEndOfMarker on a shared source — they already reuse the source, with heterogeneous asserts per position; {caretN} there is 0 reduction and messier.
$ / direct-column low-level marker tests (IdentifierIsland, QuickParse, …) — $ is more compact than {caret}; migrating adds indirection + a ±1 column risk for no reduction.

Should we convert them all, so more tests could use the same infrastructure and the extra helpers could be removed?

[T-Gro]

🤖 This review was generated by AI (@expert-reviewer agent). Findings may contain inaccuracies — please verify independently.

Assessment: no significant issues found — this is a clean, behaviour-preserving test refactoring.

Verified the two highest-risk migrations line by line:

• Symbols Nested copy-and-update: the 8 {caretN} markers map 1:1, in order, to the 8 original per-Fact (name, ((line,col),(line,col))) cases. The let nestedFunc source line is byte-identical, so the hardcoded ranges still match the token positions. List.iter2 over equal-length lists also guards against a miscounted marker (a length mismatch would throw and fail the test).
• CurriedArguments.Regression (and the other collapsed Completion sources): extractOrderedMarkedSources (Checker.fs) sorts markers by Id and re-emits each source with only its own single caret, so each generated source is identical to the corresponding original InlineData single-caret source; the expected-names list preserves the original per-case expectations.

The remaining changes just drop now-redundant (*loc-N*) anchor comments and unused open System, with assertions untouched. Confirmed SourceContext.extractOrderedMarkedSources exists and behaves as assumed. No correctness/security/perf/error-handling concerns apply to this change.