fix(auth)!: change timestamp to ms since Unix epoch for effective replay protection

[VasilevNStas]

Description

Changes Auth.timestamp from uint32 (minutes since project epoch) to uint64 timestamp_ms (milliseconds since Unix epoch) for effective per-message replay protection.

Problem

The previous uint32 timestamp used minute granularity. Every Auth sent within the same 60-second window carried the identical value. With the ±1 minute acceptance window, a captured frame could be replayed for up to 119 seconds. This is not a nonce — it's a coarse time window.

Fix

Before	After
uint32 timestamp = 2	uint64 timestamp_ms = 2
Minutes since 2024-01-01 UTC	Milliseconds since Unix epoch
±1 minute acceptance	~1–2 second skew window
HMAC input: 4 bytes LE/BE undefined	HMAC input: 8 bytes big-endian

The big-endian serialization is now explicitly specified in the HMAC comment.

Migration

Breaking wire change — the Auth frame format changes. All Nodes and clients must update simultaneously:

• Auth field 2 wire type changes (varint fits within 1–10 bytes)
• HMAC input length changes from 4 to 8 bytes
• Generated code symbol changes from timestamp to timestamp_ms

Closes #29

[VasilevNStas]

@yegor256 @davvd plz review this PR

[VasilevNStas]

@davvd conflicts resolved.

[davvd]

@VasilevNStas we are against compound names, only solid nouns

[VasilevNStas]

@davvd Thank you for the review note about compound names — understood, I will keep field names as solid nouns going forward.

Updating the branch now:

1. Renaming timestamp_ms back to timestamp (solid noun)
2. Rebuilding the branch on top of current master to remove unrelated diff (the old branch had drifted and picked up reverts of upstream changes)
3. Only change: uint32 timestamp = 2; → uint64 timestamp = 2; with updated comments for ms-since-Unix-epoch semantics and big-endian HMAC serialization

[VasilevNStas]

@davvd plz re-review