Reject sample and ancestor IDs equal to num_nodes in link_ancestors (fixes #3484)

[asystemoffields]

Description

Fixes #3484

ancestor_mapper_init_samples and ancestor_mapper_init_ancestors validated node IDs with > nodes.num_rows instead of >= nodes.num_rows, so a sample or ancestor ID equal to the number of nodes passed the check and was then used to index is_sample, is_ancestor and the ancestor map arrays, all allocated with exactly num_nodes entries. A standalone C program calling tsk_table_collection_link_ancestors with such an ID on a three-node table segfaults; through the Python API, tables.link_ancestors(samples=[0, 1], ancestors=[num_nodes]) on a two-sample msprime tree sequence returned normally.

• c/tskit/tables.c: >= in both checks, so these IDs return TSK_ERR_NODE_OUT_OF_BOUNDS like the other node bounds checks in the file.
• c/tests/test_tables.c: test_link_ancestors_input_errors gains the one-past-the-end sample and ancestor cases. On the current code these two assertions happen to pass on the seven-node test table (the out-of-bounds accesses land in allocator padding and the run completes), so the C test documents the behaviour rather than catching the regression; the Python test below does fail without the fix.
• python/tests/test_python_c.py: two cases added to test_link_ancestors_bad_args (fail on main, pass with the fix). test_link_ancestors in the same file asked for ancestor 3 of a tree sequence with nodes 0–2 and only passed because of this bug; it now asks for the root, node 2.
• Changelog entries in c/CHANGELOG.rst and python/CHANGELOG.rst; the :pr: numbers will be filled in once this PR exists.

Testing here: ninja -C build test_tables && ./build/test_tables (CUnit built from source locally): 34208 asserts, 0 failures; pytest python/tests/test_python_c.py python/tests/test_topology.py python/tests/test_highlevel.py python/tests/test_immutable_table_collection.py -k link_ancestors: 110 passed, 6 skipped; clang-format reports no changes.

PR Checklist:

• Tests that fully cover new/changed functionality.
• Documentation including tutorial content if appropriate. (no documented behaviour changes)
• Changelogs, if there are API changes.

Prepared with Claude Code (Claude Fable 5.1) and submitted on the account owner's authorisation; the runs above were done as described.

🤖 Generated with Claude Code

[hyanwong]

I think this is the same as #3489, which is waiting to be merged. But thanks for the extra spot @asystemoffields

[hyanwong]

Closed as dupe of #3489