fix(liteclient): write both magics in auth-complete payload (#463)

[chiliec]

What

Closes #463.

sendAuthComplete (liteclient/connection.go) built the tcp.authentificationComplete payload like this:

payload := make([]byte, 4)
binary.LittleEndian.PutUint32(payload, magicTcpAuthentificationComplete)
binary.LittleEndian.PutUint32(payload, magicPubKey)

Both PutUint32 calls write to offset 0 of the same 4-byte buffer, so the second write (magicPubKey) immediately overwrites the first (magicTcpAuthentificationComplete). The auth-complete message is therefore sent with the wrong type prefix and is 4 bytes short.

Cause

The TL schema is:

tcp.authentificationComplete key:PublicKey signature:bytes = tcp.Message
pub.ed25519 key:int256 = PublicKey

so the serialized layout must be two distinct magics — the message-type magic followed by the PublicKey constructor magic — then the 32-byte key and the length-prefixed signature. The buffer needs 8 bytes for the two magics, not 4.

Fix

Write the two magics into distinct 4-byte slots, and extract the payload construction into a pure buildAuthCompletePayload(pubKey, signature) helper so the byte layout is unit-testable without a live connection:

payload := make([]byte, 8)
binary.LittleEndian.PutUint32(payload[:4], magicTcpAuthentificationComplete)
binary.LittleEndian.PutUint32(payload[4:8], magicPubKey)

Tests

Added liteclient/auth_complete_test.go, which asserts the exact byte layout: message magic at [0:4], pubkey magic at [4:8], the public key at offset 8, the length-prefixed signature after it, and 4-byte alignment.

Verified genuine RED→GREEN: with the original construction restored, the test fails with message magic: got 0x4813b4c6, want 0xf7ad9ea6 — i.e. it catches the exact overwrite; with the fix it passes.

Validation (real results, Go 1.24.5)

• go test ./liteclient/ -run TestBuildAuthCompletePayloadLayout: PASS
• go build ./...: success
• gofmt -l and go vet ./liteclient/: clean

Note: the package's client_test.go tests (e.g. TestGeneratedMethod3) connect to a live lite server and fail in a sandbox without outbound network — I confirmed they fail identically on a clean master checkout, so that is unrelated to this change.

First-time contributor here — happy to adjust the helper's name/placement or the test if you'd prefer a different approach.

[mr-tron]

lgtm