[johnsaigle]

Adds regression tests for the Solana (SVM) watcher. #4931 is the EVM counterpart.

What the tests check

TestReplayGeneratedBundles replays 355 recorded transactions: 235 synthetic cases and 120 from mainnet. Each transaction runs through four watcher paths. The fixtures name them reobservation, observation, polling and account. The test compares the digest that each path emits for guardians to sign with a baseline in testdata/.

The baseline records what the watcher does today for these inputs. It does not prove that the watcher is correct, and it does not cover transactions outside this set. A change to the baseline needs a careful review of the diff.

The testgen package generates the synthetic data and fetches the mainnet data. testgen/README.md describes how to regenerate the fixtures.

Production code changes

The watcher logic is the same as on main. The PR makes two mechanical edits in client.go and reobserve.go. Most of the changed lines only move code that is already there.

• The solanaRPCClient interface replaces *rpc.Client. Tests supply a mock. The interface lists only methods that the watcher already calls.
• The goroutine bodies become named methods, the pattern that EVM Watcher Divergence Tests #4931 uses for the EVM watcher. Where the watcher used an inline body, it now starts a named method. The methods are runDataPump, runWatcher, runSlotFetcher, runRetryFetchBlock, runDelayedFetchBlock, runInitialFetchMessageAccount and runRetryFetchMessageAccount. Each body keeps its code. The moved code keeps the channelcheck edits from main.

The PR leaves shim.go, tx_for_addr.go, the Solana transaction v1 code (#4961) and the borsh code (#4892) untouched.

Tests of the moved code

watcher_loops_test.go tests runWatcher and runDataPump with a mock RPC client and a local websocket server.

The runWatcher tests cover a stop on cancel, the slot update on each tick, and a bad observation request. They also cover a GetSlot error and a full error channel. The runDataPump tests check that messages arrive. They also check that the pump stops on cancel when its consumer has gone.

Two tests guard the channelcheck edits from main: the full error channel test and the stuck consumer test. Each one fails if the old plain channel send returns.

Checks

• The 355 replay bundles also pass against the unmodified main logic with only the interface change. A scratch worktree confirmed this, so the baseline does not depend on the moved code.
• make lint and go test -race pass for the Solana packages.
• The CodeQL rules from codeql/ return no results for the Solana watcher.

Not covered

• The tests use real timers. No test runs the retry paths, which wait 5 seconds between attempts.
• The mock RPC client never fails on purpose. It has no timeouts or partial responses.
• The tests use no random or adversarial inputs, and they do not explore goroutine interleavings.
• testgen live needs a network RPC. CI does not run it.