Poll for block progress and retry toolchain installs

[nikolap-dat]

Summary

Two transient conditions failed real program deploys today with nothing wrong on the chain:

1. Health probe. arch-kit health sampled the block height twice, 2 s apart, and failed when it had not moved. A healthy chain still has multi-second gaps between consecutive blocks: 3.5 s between heights 48004 and 48005 on internal-b while it averaged ~9 blocks/s. prop-amm's deploy (run 37778455694) landed its single probe in that gap: blocks are not progressing … height changed from 48005 to 48005 over 2s.
   The check now polls the height once a second for up to 20 s and succeeds as soon as it moves; observation_window reports the time that took, and a sampling error still ends the check immediately. src/client.rs, with unit tests for the polling helper; README updated.

2. Toolchain install. The Anza installer and the platform-tools download behind cargo build-sbf both go through the GitHub release API, which intermittently returns a body the installer cannot parse. ausd-bridge's testnet deploy (run 37786560414) died on Failed to install platform-tools: Failed to parse Github response: error decoding response body.
   The reusable deploy-program.yml now installs the toolchain and prefetches platform-tools (cargo-build-sbf --install-only) with three attempts, installs arch-kit with three attempts, and probes health up to five times (covers a consumer pinned to an older arch-kit, or a node mid-restart).

Version bumped to 0.1.9 so the CLI change can be published; consumers pinned to the reusable workflow pick up the retries by moving their ref to the merge commit. The per-repo adapters (ausd-bridge, btc-wrapper, CLAMM, arch-vaults, arch-lend) already retry the health probe but install the toolchain in one attempt; that's a separate small change in each repo.

Test plan

• cargo fmt --all -- --check, cargo test --locked --all-targets (109 passed), cargo clippy --locked --all-targets -- -D warnings
• Workflow YAML parses
• Publish 0.1.9 and bump an adapter to it: arch-kit health against internal-b reports a sub-second observation window
• prop-amm pin moved to the merge commit: a real deploy passes the install and health steps

🤖 Generated with Claude Code

---

^{Need help on this PR? Tag @codesmith-bot with what you need. Autofix is disabled.}

[itsarchnw]

ARCH_KIT_VERSION in deploy-program.yml is still 0.1.8. The publish workflow checks that against Cargo.toml and exits 1 on mismatch, so publishing 0.1.9 will fail with "deploy-program.yml installs 0.1.8, expected 0.1.9".

If it merges as-is the reusable workflow also keeps installing 0.1.8, so consumers pick up the retries but not the health polling change.

Needs the env bumped to 0.1.9.

[itsarchnw]

Versions line up now and the publish check will pass. LGTM.