Validate abstract vector array instances

[quinnj]

Views, ranges, and bit vectors are rejected by type="array" even though the schema generator and array keywords already support abstract vectors. Accept AbstractVector instances while retaining existing concrete-array behavior.

The shared additional-items loop now iterates tail values and reports their logical JSON positions. This also fixes the existing bounds errors for offset vectors with tuple items and schema-valued additionalItems. No new exports or dependencies.

Validation:

• All 73 public controls pass on Julia 1.9.4 and 1.13.1. The original 58-control reproduction passes 34/fails 24 on 1.9.4 and passes 32/fails 26 on 1.13.1; the difference is two error-order checks on already unsupported inputs.
• Both full package suites pass 5,940 assertions with bounds checks; all tested source hashes remain unchanged.
• Controls cover generated schemas, views/ranges/bit vectors, item/length/contains/uniqueness constraints, empty/nullable/nested values, offset-vector tuple tails and error pointers, and existing concrete-matrix type acceptance. The independent offset reproduction passes on Julia 1.9.4 and 1.13.0 after the shared-loop fix.
• JuliaFormatter 2.14.0 with the repository configuration, git diff --check, and a complete independent private review pass. Hosted CI and coverage processing are pending.

Co-authored by Codex

AI disclosure: This work was prepared with assistance from OpenAI Codex.

Validation after publication: all seven exact-head checks passed, and all six correct-repository coverage uploads were accepted and processed (96.46%). The ordinary merge has the same tested Git tree; master also passed all seven checks and all six accepted and processed coverage reports (96.46%).

[quinnj]

Merged the array-instance and offset-indexing fixes. Concrete arrays retain their existing behavior; one-dimensional views, ranges, and bit vectors now receive array validation, and tuple tail checks use logical JSON positions for offset-indexed vectors.

All 73 regression assertions and both full 5,940-assertion suites pass on Julia 1.9.4 and 1.13.1. The exact published head and identical merge tree each passed all seven hosted checks and all six accepted, processed coverage uploads (96.46%).

AI disclosure: This work was prepared with assistance from OpenAI Codex.