fix: use original directory name for temp URI lookup in TreeBuilder

[ponsde]

Problem

TreeBuilder._finalize_from_temp() applies VikingURI.sanitize_segment() to
the temp document directory name (e.g. __test__ → _test_), then uses the
sanitized name to construct temp_doc_uri. However, the actual temp directory
on disk retains its original (unsanitized) name, causing FileNotFoundError
during _move_temp_to_dest() when the name contains consecutive underscores
or other characters that sanitize_segment() modifies.

Fix

Use the original directory name for temp_doc_uri (locating the temp
directory), while continuing to use the sanitized name for candidate_uri
(the final destination URI). Added debug log when sanitization changes the name.

Changes (1 file, +5 −2)

• openviking/parse/tree_builder.py: Split doc_dirs[0]["name"] into
  original_name (for temp lookup) and doc_name (sanitized, for final URI).

Test plan

• add_resource with filenames containing __, leading/trailing underscores
• add_resource with normal filenames (no sanitization change) still works
• Verify final URI uses sanitized name, temp lookup uses original name
• Sanitized name conflicts with existing URI: verify _resolve_unique_uri() still handles renaming correctly

[ponsde]

The test-lite failure appears to be a pre-existing infrastructure issue unrelated to this PR — the main branch CI also shows widespread test failures today (see run #22442035868), likely related to the recent agfs/binding changes. Our fix only touches lines 130-131 of tree_builder.py (temp URI lookup), which is not exercised by the quick-start lite test.