Replace expanded sequence arrays with a compact deterministic representation

[JackWilb]

Summary

Large studies currently persist and transfer 1,000 fully expanded participant sequences. In library-calvi, the sequence artifact is approximately 3.35 MiB. Even after duplicate downloads are removed by #1176/PR #1311 and publication is moved out of the interaction barrier by #1319, this representation remains expensive to generate, upload, download, parse, and retain.

Replace the expanded artifact with a compact deterministic descriptor while preserving every existing assignment and randomization behavior.

Design direction

Persist a versioned descriptor containing the minimum information needed to reproduce an assigned sequence, such as:

• Config hash.
• Deterministic seed.
• Sequence-generation algorithm version.
• Compact permutation/order data only where it cannot be derived from the seed.

A descriptor must produce identical results across supported browsers and storage engines. Algorithm changes must create a new version rather than silently changing sequences for an existing descriptor.

This work should build on the config-hash build coordination in #1319. Existing full sequence arrays remain a supported legacy format.

Requirements

• Define a versioned persisted schema for compact sequence descriptors.
• Make sequence generation deterministic for a given config hash, seed, and algorithm version.
• Derive only the assigned row when possible rather than expanding every participant sequence during startup.
• Preserve Latin-square ordering, random ordering, URL conditions, stages, nested/dynamic blocks, rejected-slot reuse, and participant-resume behavior.
• Continue reading existing expanded sequence arrays without manual migration.
• Preserve access to the exact config/sequence representation used by an existing participant.
• Support Firebase and Supabase; preserve LocalStorage behavior.
• Detect unsupported or corrupt descriptor versions and fail with a recoverable, actionable error.
• Do not add a dependency unless separately approved.

Acceptance criteria

• A newly published study stores a compact versioned descriptor instead of 1,000 expanded sequences.
• The descriptor is materially smaller than the current expanded artifact for library-calvi, with before/after byte sizes recorded.
• The same descriptor produces byte-equivalent assigned sequences across repeated runs and supported browsers.
• Concurrent clients resolving the same descriptor cannot produce different sequences.
• Existing expanded arrays and participant sessions continue to load without migration.
• Tests compare legacy and compact output for Latin-square, random, conditional, staged, and dynamic study configurations.
• Tests lock algorithm-version behavior so future generator changes cannot alter existing participant sequences.
• Browser profiling records download, parse/generation, and time-to-interactive changes for library-calvi.
• yarn unittest --run, yarn typecheck, yarn lint, and yarn build pass.

Out of scope

• Changing study balancing or randomization semantics.
• Removing compatibility with existing expanded arrays.
• General application bundle optimization.