Review: fix: trim empty top-level else branches

Verdict: Approve

What the PR does

Fixes issue #497: when a false-branch destination was already visited via the true branch, traverseFlowUntilMerge returned immediately at the visited[currentID] guard (cmd_microflows_show_helpers.go:618) without appending anything. The old check falseFlow.DestinationID != mergeID only excluded false branches that pointed directly at the merge — it didn't account for branches pointing at a non-merge node that had already been emitted. The result was an else header with an empty body.

The fix switches to a tentative-emit-then-retract pattern: append the else header, run the traversal, then check if anything was added. If len(*lines) == elseLineIdx+1 (only the else header itself was appended), trim it back.

No blockers

Correctness notes

• len(*lines) == elseLineIdx+1 is the right sentinel — it fires only when nothing was appended after the else header. If the traversal emits any line (including an inner if ... end if; with no statements), the outer else is correctly kept.
• The previously-working case (falseFlow.DestinationID == mergeID) still works identically: traverseFlowUntilMerge stops at the merge boundary without appending, then the else is retroactively removed.
• visitedFalseBranch is a copy of visited at the time of the false branch traversal, so the shared-node scenario the test covers is correctly detected.
• The PR summary says "at top level" but the change is in traverseFlow which handles all nesting depths — the description is slightly misleading but the logic is correct at all levels.

Positive notes

• Tentative-emit-then-retract is cleaner than trying to predict traversal output ahead of time — it handles all "empty false branch" causes with one check
• Test models the exact scenario: outer split → true → inner split → true → shared continuation; outer false branch jumps directly to the same shared node already visited via true
• Clean PR, no bundled unrelated changes