refactor(starknet): replace BlockStatus string with typed enum

[0xsamalt]

• Introduces BlockStatus as a typed uint8 enum in starknet/block.go with constants BlockPending, BlockAcceptedOnL1, BlockAcceptedOnL2, BlockRejected, and BlockPreConfirmed
• Adds UnmarshalText so the feeder JSON response deserialises directly into the enum and invalid status strings now return an error
• Updates all comparison sites and test assertions to use the typed constants

[Ehsan-saradar]

Nice one, typed status is much better than a bare string 👍

One thing I'd double-check first though: UnmarshalText now errors on any status that isn't one of the five, and since blocks are decoded straight through json.Unmarshal, an unknown status fails the whole block decode — not just that field. With the old string we accepted anything and only really cared about PENDING / PRE_CONFIRMED.

So if the feeder ever sends something not in the list (pretty sure ABORTED / REVERTED can show up on blocks), we'd fail to read that block and sync would get stuck on it.

Wdyt about mapping unknown values to a BlockUnknown instead of erroring? Keeps it harmless like before, since we only check for PENDING / PRE_CONFIRMED anyway. Or if we'd rather keep it strict, probably worth confirming the list covers everything the gateway can actually send.

[0xsamalt]

Checked this against both the spec and Juno's own fixtures. BLOCK_STATUS is a closed enum (PENDING / ACCEPTED_ON_L2 / ACCEPTED_ON_L1 / REJECTED, plus PRE_CONFIRMED since RPC 0.9) and ABORTED/REVERTED aren't block statuses, they're TXN_EXECUTION_STATUS values scoped to individual transaction receipts.

Grepped our own feeder testdata to be sure this isn't just spec theory: REVERTED shows up in exactly one fixture, and it's a transaction file, never a block file. Across every block and state-update fixture we've captured from the real feeder gateway, block-level status has only ever been PENDING/ACCEPTED_ON_L2/ACCEPTED_ON_L1/REJECTED/PRE_CONFIRMED, no evidence the feeder has ever sent anything else at the block level.

So I don't think there's a real path that hits the default error case. Given that, I'd rather keep it strict if the feeder ever does send something unexpected here, that's a signal of an actual protocol change we want to know about loudly, not something we want to silently coerce into a BlockUnknown fallback and risk masking. Happy to add the fallback anyway if you'd still prefer defensive handling, but wanted to share what I found first.

[Ehsan-saradar]

Fair enough — the fixture check convinced me, let's keep it strict. Only nit: could the UnmarshalText error include the offending value (e.g. unknown block status %q)? Makes a future protocol change point straight at itself. 👍

[0xsamalt]

Already included... the default case returns fmt.Errorf("unknown BlockStatus %q", str), so the error message carries the offending value.

[0xsamalt]

Rebased onto main to resolve conflicts. A few refactors landed on main since this opened (subscription-cache→deduper, feeder client value-return change, pending-endpoint removal) that touched the same code paths.

No change to the BlockStatus type or its logic; the rebase also required updating a handful of test call sites (clients/feeder/feeder_test.go, sync/preconfirmed/chain_storage,poller_test.go, rpc/v9|v10/transaction_test.go) that still used the raw "PRE_CONFIRMED" string instead of the typed constant, since those tests were touched by other PRs after this one opened.

Verified with go build/go vet/go test across the affected packages.