Recover from mismatched loop terminators with quick fixes

[chrisdp]

Summary

When a user writes a common loop-terminator mistake, the parser used to emit two cascading diagnostics that didn't connect the user's intent to the actual problem. This PR adds error recovery in three loop-parsing paths and a single parameterized diagnostic with a structured data payload that drives the quick fixes.

Before

while n <= 3
    n = n + 1
next

emitted:

• Unexpected token 'next'
• Could not find matching 'end while'

After

Same input emits one diagnostic:

• Expected 'end while' but found 'next' (code 1147)

with a quick fix Convert 'next' to 'end while'.

Cases handled

Input	Diagnostic message	Quick fix(es)
while ... next	Expected 'end while' but found 'next'	Convert 'next' to 'end while'
for ... end while	Expected 'end for' or 'next' but found 'end while'	Convert 'end while' to 'end for' (preferred), Convert 'end while' to 'next'
for each ... end while	same as for	same two fixes

When the same mistake appears multiple times in a file, fix-all variants are also offered.

Implementation

• New mismatchedEndingToken(expected: string[], found: string) diagnostic (code 1147). The expected array is preferred-first, and both fields are kept on diagnostic.data so consumers can build fixes mechanically rather than dispatching on per-pair codes.
• whileStatement accepts Next as a recovery terminator; consumes it on the bogus path and stores it on WhileStatement.tokens.endWhile so quick fixes can target the token's range. Calls mismatchedEndingToken(['end while'], 'next').
• forStatement and forEachStatement accept EndWhile as a recovery terminator; same shape. They call mismatchedEndingToken(['end for', 'next'], 'end while'). forEachStatement also refactored from "throw on missing terminator" to recover-and-continue.
• CodeActionsProcessor.suggestMismatchedEndingTokenQuickFixes reads expected[]/found from the diagnostic and emits one fix per legal terminator (first entry is isPreferred). No per-pair handler.
• New isDiagnosticOfType(diagnostic, key) type guard refines BsDiagnostic to the concrete DiagnosticMessageType<K> when its code matches, used by the code-action dispatcher.

Test plan

• npm test — 2838 passing, 0 failing
• npm run lint — clean
• Parser specs assert the new diagnostic code and data: { expected, found } payload for while, for, and for each
• Code-action specs cover single-fix, two-fix (with preferred), and fix-all variants