refactor(starknet): change BlockIdentifier from string to typed uint64

[0xsamalt]

• Introduces starknet.BlockIdentifier as a named uint64 type with custom UnmarshalJSON that handles both hex string (ex: "0x...") and bare integers.
• Changes PreConfirmedDeltaUpdate.BlockIdentifier and PreConfirmedBlock.BlockIdentifier from string to the new BlockIdentifier type.
• Changes pending.PreConfirmed.BlockIdentifier from string to uint64 (raw, to avoid a circular import from core/pending → starknet).
• Updates all mocks, adapters, and tests to match the new types.

[Ehsan-saradar]

Heads up — this doesn't compile yet. A few PreConfirmedBlock* call sites still pass "" for the identifier, which no longer type-checks now that it's starknet.BlockIdentifier. Both the lint and test jobs fail on these three (everything else builds fine and the rest of the suite is green):

• rpc/v9/transaction_test.go:473 — PreConfirmedBlockWithIdentifier(..., "", 0)
• rpc/v9/transaction_test.go:1793 — PreConfirmedBlockByNumber(..., "", 0)
• rpc/v10/transaction_test.go:600 — PreConfirmedBlockWithIdentifier(..., "", 0)

Changing the "" to 0 in those three spots does it 👍

[0xsamalt]

• rpc/v9/transaction_test.go:473 — PreConfirmedBlockWithIdentifier(..., "", 0)
• rpc/v9/transaction_test.go:1793 — PreConfirmedBlockByNumber(..., "", 0)
• rpc/v10/transaction_test.go:600 — PreConfirmedBlockWithIdentifier(..., "", 0)

Fixed, tests passing locally on rpc/v9 and rpc/v10

[Ehsan-saradar]

Perfect, thanks for the quick fix — builds clean now. 👍

[RafaelGranza]

Hey @0xsamalt, thanks for the PR!

Could you resolve the conflicts first? And could you help me understand the goal here?

It looks cosmetic to me, and it touches a very delicate spot.

Is it tied to any issue? Do you think it improves performance? If so, a benchmark would help.

[0xsamalt]

Thanks @RafaelGranza!

• Yes I am on it.. will be resolving all the conflicts
• Also yes it is tired to the issue Change BlockIdentifier and BlockStatus types in the starknet pkg #3767
• goal is mainly type safety, not perf. Raw string BlockIdentifier let bad input slip past until it hit deep logic; typed uint64 + custom UnmarshalJSON catches it at the parse boundary instead.

[0xsamalt]

Rebased onto the latest main (455b0dd) to resolve conflicts from the multi-block pre-confirmed chain work, the subscription-cache to deduper refactor, and the pending-endpoint removal that landed since this branch was opened.

Squashed down to a single commit the follow-up fix commit's changes were already folded in while resolving conflicts, so git dropped it as redundant.

Besides mechanical conflict resolution, this also:

• Propagates the BlockIdentifier type change through code that landed on main after this PR branched (clients/feeder.Reader, starknetdata.StarknetData, sync.DataSource, sync/preconfirmed.DataSource, rpc/rpccore.PreConfirmedDeduper), which still had stray string-typed identifier params.
• Fixes a real regression surfaced by testing during the rebase: DecodePreConfirmedUpdate was decoding block_identifier before checking changed, so a malformed identifier on a changed:false (NoChange) response broke decoding, even though NoChange is documented to ignore all other fields. Now the discriminator is checked first.

Full build, go vet, and test suite pass locally.