[VOTE] Stabilize field IDs across schema evolution

[Xuanwo]

Formal vote on PR #8658.

This proposal makes field IDs stable and non-reusable within a dataset branch ancestry by persisting a monotonic allocation high-water mark. New datasets enable the contract by default with required reader and writer feature flags; existing legacy datasets remain unchanged until explicitly migrated.

Please vote with:

• +1 approve
• 0 abstain / no strong opinion
• -1 do not approve (please include the blocking concern)

Please keep detailed design discussion in the PR.