fix: rejoin empty error handlers inside branch bodies

[hjotha]

Closes #495.

Summary

• routes pending empty no-output custom error handlers when IF branch bodies connect one statement to the next
• applies the same fix to enum-split branch bodies
• keeps output-dependent custom handlers on the existing skip-var path
• adds a synthetic IF-branch regression test

Validation

• go test ./mdl/executor -run 'EmptyNoOutputHandler.*Branch|EmptyNoOutputHandlerRejoins|OutputHandlerInReturningBranch' -v
• make build
• make lint-go
• make test

[ako]

Review: fix: rejoin empty error handlers inside branch bodies

Verdict: Approve with minor note

What the PR does

Fixes issue #495: when a CALL MICROFLOW with an empty no-output custom error handler appears inside an IF branch body (then or else) or an enum-split branch body, the pending error handler flow wasn't wired to the next statement in that branch. The fix adds the same rejoin call that the top-level flow loop already performs.

No blockers

Minor: outer guard is redundant, creates inconsistency with the top-level pattern

The three new blocks read:

if fb.emptyErrorHandlerFrom == lastThenID {
    fb.addPendingErrorHandlerFlowForStatement(...)
}

But addPendingErrorHandlerFlowForStatement → addPendingErrorHandlerFlowForState already has this guard internally:

if state.emptyFrom != "" {
    if state.emptyFrom != originID {
        return state  // no-op
    }

Calling it when there's no pending handler, or when the handler is from a different origin, is already a safe no-op. The top-level loop in cmd_microflows_builder_graph.go:98 calls addPendingErrorHandlerFlowForStatement unconditionally — no guard. The branch-body version's outer if is functionally correct but redundant, and creates a subtle inconsistency that makes readers wonder why the two call sites look different.

Could drop the outer check to mirror the top-level pattern. Not a blocker — the code is correct either way.

Positive notes

• Clean, focused fix — 3 lines of logic added in each of 3 places, matching the existing top-level pattern
• br.body[j+1:] and s.ThenBody[i+1:] are safe when j/i is the last index (empty slice is valid in Go)
• New test directly reproduces the failure: IF-branch with no-output handler followed by a continuation activity, asserts the error path reaches the continuation
• No bundled unrelated changes