test(pricefeedwatcher): assert only one watch loop is started

[rickstaa]

Follow-up to #4031, targeting ja/golang-1.27 so it lands with the Go 1.27 bump.

#4031 removes the reflect-based cancel func comparison from TestPriceFeedWatcher_Subscribe. That removal is correct, but it leaves the "only 1 watch loop gets started" comment describing something nothing checks. This adds a test that actually checks it.

Why the old assertion could not work

It compared reflect.ValueOf(w.cancelWatch).Pointer() across subscriptions. For a func value that returns the address of the compiled function body, not the identity of the closure. Every cancelWatch comes from context.WithCancel, so all of them share one code address:

c1=0x4933a0 c2=0x4933a0  equal=true

Two unrelated cancel funcs, guaranteed equal. The assertion passed whether ensureWatchLocked assigned once or five times, so it could not detect the bug it was written for. reflect's docs say as much: for a Func the result "is not necessarily enough to identify a single function uniquely".

What this asserts instead

A duplicate watch loop is observable: each loop gets its own ticker and polls w.priceFeed once per priceUpdatePeriod. So N loops means N FetchPriceData calls per tick.

priceUpdatePeriod is 1 hour, which is why this was not testable before. testing/synctest (already used elsewhere in #4031) makes the hour instant, puts newTruncatedTicker on a deterministic boundary, and synctest.Wait() blocks until every bubble goroutine is idle. No sleeps, no flake.

Verified both directions

• Against the real code: passes.
• With the if w.cancelWatch != nil { return } guard in ensureWatchLocked removed: fails with Expected number of calls (1) does not match the actual number of calls (5).

🤖 Generated with Claude Code