fix(http): send market orders as FrontendMarket, not GTC

[changshenhan]

Summary

Two related price-handling fixes in the trading path.

1. market_open sends real market orders (was sending GTC limits)

market_open() is documented as using Hyperliquid's native FrontendMarket order type ("fills immediately up to the provided worst acceptable limit price"), and the example examples/hypercore/market_order.rs says the same — but the code built the order with TimeInForce::Gtc:

order_type: OrderTypePlacement::Limit {
    tif: TimeInForce::Gtc,   // ← rest as a passive limit
},

TimeInForce::FrontendMarket has always existed (src/hypercore/types/mod.rs) but was never actually used anywhere. The practical impact on live trading:

• A market buy whose limit_px (worst acceptable) is below the current ask rests as a passive GTC limit instead of filling — the caller believes they're filled, their position isn't.
• A market sell with limit_px above the current bid never fills unless price crosses it.

FIX: Gtc → FrontendMarket. This is the Hyperliquid-protocol market order (a Limit with tif: FrontendMarket), matching the frontend semantics the method documents.

2. PriceTick::tick_for correct at powers of ten

tick_for's doc algorithm is sig_figs = floor(log10(price)) + 1, but the code used ceil(log10(price)). The two agree for every non-power-of-ten price and disagree exactly at powers of ten, where ceil undercounts the integer digits by one and returns a tick ten times too fine:

price	documented	actual (before)	actual (after)
1000	0.1	0.01 ❌	0.1 ✅
100	0.01	0.001 ❌	0.01 ✅
1	0.0001	0.00001 ❌	0.0001 ✅

Visible on any market with max_decimals ≥ the computed value (e.g. a perp/spot with low sz_decimals); round/round_by_side inherit it. No existing test covered a power-of-ten price — added power_of_ten_prices_keep_five_sig_figs.

Testing

• New regression test for power-of-ten prices.
• cargo test --lib: 149 passed, 0 failed.
• No new clippy warnings (23 pre-existing warnings on main are unchanged).