Minor code TODOs and Postgres polish

[ihistand]

Small cleanups / known limitations gathered from in-code TODOs and CLAUDE.md nice-to-haves. None are urgent; good incremental work.

• Running tests on incremental datasets is unsupported (core/actions/test.ts)
• Postgres adapter truncates large values — should use Long (cli/api/utils/postgres.ts)
• data_preparation: incremental conditions not in compiled output (TODOs in core/actions/data_preparation.ts)
• cli/index.ts is not type-checked (migrate to typed yargs)
• Test results capped at 1 MB (cli/api/commands/test.ts)
• Postgres partitioning gaps: storage options on partitioned parents, sub-partitioning, opclasses on partitioned indexes

[ihistand]

Re-triaged. This was a mislabeled grab-bag — the items are unrelated and mostly not "minor." Closing in favor of focused issues + won't-fix decisions.

Split into focused issues:

• Migrate cli/index.ts to typed yargs #18 — migrate cli/index.ts to typed yargs (tech-debt)
• Paginate unit-test results (remove the 1 MB cap) #19 — paginate unit-test results (remove the 1 MB cap)
• Support unit tests on incremental datasets #20 — support unit tests on incremental datasets
• Postgres partitioning enhancements: sub-partitioning + tablespace on partitioned parents #21 — Postgres partitioning enhancements (sub-partitioning; tablespace on partitioned parents)

Won't-fix (deliberate / out of scope):

• BIGINT→Number truncation (cli/api/utils/postgres.ts) — intentional and consistent with the other adapters; switching to Long/BigInt changes result types and breaks JSON serialization. Not worth the blast radius.
• data_preparation incremental conditions — data_preparation is a niche BigQuery action; low priority for a Postgres/Supabase-focused fork.

Not actually a gap (correction): "opclasses/indexes on partitioned tables" already work — createIndexes() runs for the partitioned path (postgres_execution_sql.ts:185). fillfactor on a partitioned parent is invalid (Postgres forbids storage params on partitioned tables).