Add HYPE (Hyperliquid) as a launch spoke

[LandynDev]

Adds SOL ↔ HYPE. Stacked on #637 — merge that first; this PR's diff is only the chain itself.

HyperEVM is a plain EVM JSON-RPC chain, so the pair is a registry row plus a binding of the shared EvmCoin. No contract, DB, or consumer changes.

allways/chains.py              CHAIN_HYPE row                       + 21
allways/assets/evm.py          HYPERLIQUID network (chain ids, RPC)  + 12
allways/assets/hyperliquid.py  class Hype(EvmCoin)                   + 12
allways/assets/__init__.py     registry row                          +  3
allways/constants.py           LAUNCH_SPOKES                         +  1

Everything else in the diff is .env.example, README, and tests.

Chain parameters

Field	Value	Why
decimals	18	Native gas coin on HyperEVM
seconds_per_block	1	~1s small blocks; transfers never need the 60s big blocks (those carry >2M gas)
min_confirmations	2	HyperBFT finalizes the block sequence on inclusion — no reorgs. 2 is a one-block margin against an endpoint serving an unsealed head
min_onchain_amount	1e14 wei	Rate-sanity floor: 0.0001 HYPE covers a 21k-gas transfer below ~4.7 gwei
replay_grace_secs	60	Hub-clock vs spoke-clock skew allowance, as on Arbitrum
networks / testnet_network	mainnet, testnet / testnet	chain id 999 / 998, checked against HYPE_NETWORK at startup

Default keyless RPC ladders ship for both networks (rpc.hyperliquid.xyz, drpc); those are capped at 100 req/min per IP, so HYPE_RPC_URLS should point at a keyed endpoint (e.g. Dwellir) in production.

The one shared-code change: evm_coin.py gains a hard ceiling on the deposit scanner's block walk. At 1s blocks the existing 5-minute lookback would be ~300 sequential eth_getBlockByNumber calls on a first scan. The bound never binds on ~12s chains, so ETH behaviour is unchanged.

Verification

• uv run pytest tests/ -q → 1108 passed; ruff format + check clean.
• tests/test_hype_provider.py (26 cases) covers network selection, the wrong-network startup rejection, inclusion vs settlement, receipt/timestamp unavailability raising rather than reading as absent, the full signed payload (chain id, EIP-1559 fees, estimated gas), send guards, and the scanner's bound + cursor parking.
• The pinned CLI contract from Share EVM coin behaviour and derive chain CLI config #637 (tests/test_cli_chain_networks.py) required a conscious update here — hype-network in the settable keys, hype: testnet in the testnet bundle, hype: mainnet in the mainnet bundle. A new chain cannot slip into the CLI unnoticed.
• Live read-only pass on both networks: startup chain-id check, tip, a real settled transfer verified through an UPPERCASED recipient with the sender pinned, underpayment / wrong-sender / unknown-tx all rejected, balance read.
• alw miner quotes --help shows --hype-price/--hype-address; alw config shows the hype-network row; alw config set --help lists mainnet | testnet — all derived, none hand-written.

Notes

• The program needs nothing: chains are opaque strings, and an unlisted chain gets the 600s default fulfillment grace, far above 2 × 1s confirmations.
• A fifth spoke dilutes the zero-volume emission floor to (1−α)/pairs, as with every added pair.
• das-allways companion: entrius/das-allways#86. Its CI drift gate stays red until this merges (it reads chains.py from test); verified green against this branch with ALLWAYS_BRANCH=feat/hype-spoke.