Make partition table cell edits deterministic (review of #144)

[alexeid]

Two changes from reviewing #144, targeted at beautitests so they land with that PR. One is a review comment on #144's code; the other is an opportunistic cleanup in a file #144 does not touch. Happy to drop the second if you'd rather keep the PR tight.

1. The new cell lookup can land on a recycled empty cell (review comment on this PR)

robot.lookup("#" + cellId).query() returns the first node with that id anywhere in the scene, unfiltered. AlignmentListInputEditor's cell factory sets the id and the ComboBox graphic on every TableCell it creates — including the recycled cells backing the empty rows below the table (updateItem only does setVisible(!empty)) — and those keep a live setOnAction(e -> action(e, COLUMN)) handler. If the lookup lands on one, the synthesized ActionEvent reaches AlignmentListInputEditor.action(), which does tableEntries.get(row.getIndex()) with an index of -1, and the intended partition is never renamed. With no match at all, the cast failed with a bare NPE instead of a useful message.

The lookup now iterates the candidates and picks the visible, non-empty cell whose TableRow index matches the requested row, throwing a descriptive AssertionError if there is none. A (robot, row, col, string) overload was added; the existing three-arg form delegates to row 0, so no caller changes were needed.

2. BeautiSimpleTest adopting the new helper (cleanup, not a defect in this PR)

BeautiSimpleTest.simpleTest renames the tree with the same clickOn(#treeModelCell) + eraseText(10) + write() sequence that setPartitionTableCell was introduced to replace, then asserts assertStateEquals("Tree.t:tree", …). It now calls setPartitionTableCell(robot, 7, "tree") instead.

To be clear about what this is: BeautiSimpleTest is byte-identical before and after #144, does not call setPartitionTableCell, and passes at the pre-#144 baseline — so it was not one of the tests failing in #143 and this is not an incomplete fix on your part. It is just consistency: one remaining copy of a pattern this PR establishes a better replacement for. If you'd prefer, I'll pull it out and raise it separately against master after #144 lands.

Verification

• mvn -pl beast-fx -am test-compile passes.
• BeautiSimpleTest passes with these changes (Tests run: 1, Failures: 0), and also passes unchanged at origin/beautitests. Neither failure mode reproduces on a single isolated run on my machine, so treat both changes as hardening against timing/scene-graph-order dependence rather than as a demonstrated red→green fix.
• I have not run BeautiDivergenceDatingTest or BeautiCLITest against this branch.

Note, not addressed here

All three BeautiCLITest bodies sit inside if (!GraphicsEnvironment.isHeadless()). Under a headless run the bodies — including the f.exists() && f.length() > 0 assertions — never execute, so those tests pass vacuously and the rewritten template is not actually exercised. Pre-existing, but worth knowing when judging whether CI verifies this PR's fix.