fix(formatting): recover from token spans outside property children

[ahfoysal]

Summary

• Fix formatting recursion guard in processChildNode to handle scanner tokens that overrun the current child node boundary in recovery scenarios.
• Add regression fourslash test for the invalid type-literal property syntax (type X = { prop: = \${abc}`xyz }) that triggered Debug Failure. False expression: Token end is child end`.

Validation

• npx hereby runtests --tests formattingTokenEndChildEnd --light=false
  • ✅ 1 passing (fourslash regression)

Fixes #63598.

[ahfoysal]

@microsoft-github-policy-service agree

[Copilot]

Pull request overview

Fixes a formatter debug assertion in processChildNode by improving the scanner/AST synchronization logic when recovery parsing produces tokens whose scanned spans exceed the current child node’s end boundary. Adds a fourslash regression test reproducing the invalid type-literal property syntax from #63598.

Changes:

• Update processChildNode’s pre-child token consumption loop to rewind the formatting scanner when the current scanned token overruns the child’s end.
• Add a fourslash regression test that formats an invalid type-literal property and asserts formatting completes without crashing.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated no comments.

File	Description
tests/cases/fourslash/formattingTokenEndChildEnd.ts	New regression test covering the recovery scenario that previously triggered the formatter assertion.
src/services/formatting/formatting.ts	Adjusts scanner rewind conditions to handle tokens spanning beyond a child node in recovery parsing, preventing Debug.assert(tokenInfo.token.end === child.end) failures.

[MartinJohns]

@ahfoysal You should read this: https://github.com/microsoft/TypeScript/blob/main/CONTRIBUTING.md