[P2] sqlite index for padrón (sub-ms RUC lookups)

[Railly]

Context

PR #3 ships streaming lookup over the 1GB padrón TXT (~1s per RUC, single scan for batch). Fine for ad-hoc use.

If batch sizes grow >1000 RUCs frequently, sub-ms lookups via sqlite would matter. Bun has built-in sqlite.

Scope

• `sunat padron index` — one-time index build after `padron sync`
• Schema: `(ruc TEXT PRIMARY KEY, razon_social, estado, condicion, ubigeo, ...)`
• `lookupRuc` and `lookupRucBatch` use sqlite when index exists, fall back to streaming when not
• `padron status` reports index presence + size

Trade-offs

• Disk: ~1GB extra (index parallel to TXT). Could replace TXT entirely with sqlite-only.
• Index build: ~30s on first run, then instant lookups
• Schema migrations as SUNAT adds columns

Success criteria

• 1000 RUC batch goes from ~1s to <50ms
• LIMITATIONS.md → padrón section reflects new option

Why P2

Streaming 1s is good enough for current single-user UX. Only matters when an agent does `for ruc in 5000_rucs`.

References

• src/sunat-rest/padron-local.ts → current streaming impl