CLI: readable wallet commands (status, balance, address, send)

[awizardxch]

A first increment for #397, following the steps Rigidity described there: convert user arguments into the sage-api types, call the RPC (several requests where a command needs them), and format the output into something more readable than JSON.

sage wallet

sage wallet status                 # wallet name and fingerprint, network, spendable balance, sync progress, receive address
sage wallet balance                # XCH, and every visible token with a balance
sage wallet address                # the current receive address
sage wallet send <address> <amount> [--fee 0.0001] [--asset <cat id>] [--memo "text"] [--yes]

--json on any of them prints the RPC responses instead. sage rpc <endpoint> '<json>' is unchanged.

How send stays safe

• Exact amounts. 0.25 is parsed digit by digit at the asset's precision: 12 for XCH, read from the wallet's unit; a CAT's own from get_token. There's no floating point. More decimals than the asset has is an error, not a rounding. Signs, exponents, separators and overflow are refused. Unit tests cover all of these.
• Preview what was built, then sign exactly that. The wallet first builds the transaction with auto_submit: false. The CLI prints what leaves the wallet, the fee and the number of coins spent, all read from the built transaction's summary rather than echoed from the arguments, so a wrong address or amount shows before anything is signed. On confirmation, the CLI signs and submits those same coin spends with sign_coin_spends. It never re-requests the send, which could select different coins than the ones shown.
• No silent signing. Without a terminal to confirm on, send refuses unless --yes is given.
• Text memos are hex-encoded the way the Send page does it. Token sends include the hint, which is the API default.

Also: the CLI no longer overflows its stack

On upstream main, a Windows debug build of sage overflows its stack on any command, sage --help included. The rpc command's clap tree (one subcommand per endpoint) and its handler future outgrow the 1 MiB main-thread stack. The CLI now runs on a 16 MiB thread, with the same size for the tokio workers. After the change, sage --help, sage wallet --help and sage rpc … all start normally.

Tests

• cargo test -p sage-cli: 5 tests for amount parsing and formatting (exact parsing, refusing to round, refusing non-amounts and overflow, formatting without trailing zeros, round trips).
• cargo clippy -p sage-cli --all-targets and cargo fmt --check are clean.
• Checked by hand: sage --help, sage wallet --help, and sage rpc get_version reaching the client (no wallet RPC was running here, so it reports the connection error instead of crashing).

Next steps for #397 (not in this PR)

More commands in the same style: offers (make, take, cancel, list), NFTs, coin management (split, combine), and transactions. Plus a shared table/text formatter if the command count grows.

🤖 Generated with Claude Code