Fix finally-region cloning losing break/continue/parentTryOp wiring

[ASDAlexander77]

Summary

Follow-up to #232, addressing the last actionable finding (item 5) from the LowerToAffineLoops.cpp/LowerToLLVM.cpp lowering-passes review (docs/LowerToAffineLoops-LowerToLLVM-review.md).

• TryOpLowering's finally region cloning — cloneRegionBefore is called twice to build the normal-exit and return-cleanup copies of finally, producing entirely new Operation* instances each time. Any tsContext side-table entry (jumps[] for break/continue, parentTryOp[] for a nested try) keyed on the original region's ops — populated by walking that region before cloning — was silently missing for both clones; only the third, inlined copy (which preserves original identity) kept correct wiring. Fixed by passing an explicit mlir::IRMapping to each cloneRegionBefore call and re-propagating matching side-table entries from each old op onto its clone via the mapping's getOperationMap().
• Bonus fix, found while building the regression test: a try nested directly inside a finally block failed to compile at all (error: failed to legalize operation 'ts.Try' that was explicitly marked illegal), independent of break/continue. Root cause: the dialect-conversion driver legalizes a cloned nested TryOp recursively while the outer TryOp's own pattern application is still on the stack, and refuses to reapply the same shared Pattern instance recursively without an explicit setHasBoundedRewriteRecursion() opt-in — which no TsPattern had. Fixed by giving TryOpLowering its own constructor that opts in (safe here: each recursive application strictly consumes one TryOp).
• Found but deliberately NOT fixed, tracked in the review doc (item 5b): break/continue inside a try body (as opposed to directly inside finally) does not run the enclosing finally block before jumping — reproduces a JIT crash on unmodified main, predates all three passes of this review. Out of scope for this change; the new test only covers break/continue placed directly inside finally, which works correctly.

Test plan

• cmake --build __build/tslang/windows-msbuild-2026-debug --config Debug -j8 — clean build
• ctest --test-dir __build/tslang/windows-msbuild-2026-debug -C Debug -j8 — 690/690 passed (JIT + AOT)
• New 00try_finally_break_continue.ts: break directly in finally, continue directly in finally, and a nested try/catch inside finally — each exercised across multiple loop iterations to catch "only the first/last clone copy is wired correctly" bugs. JIT + AOT green.