demos/counter: improve atomic use and parse duration as int64

[Jaana Dogan (rakyll)]

Changes

• Use atomic.Uint64: Change requestCount from uint64 to atomic.Uint64.
• Direct int64 parsing: Use strconv.ParseInt directly without intermediary int` casts.

[Dmitry Berkovich (dberkov)]

The title, body, and commit message all describe a third change — "parse duration as int64" via strconv.ParseInt — but it is not in the diff: /set-sigterm-sleep still uses strconv.Atoi plus an int64(d) cast, byte-identical to main.

As is, the commit message would land on main describing a change that never happened (per AGENTS.md, commit messages should describe the change and stand on their own). Please either implement the strconv.ParseInt change or strip that claim from the title/body/commit message.

[Jaana Dogan (rakyll)]

PTAL

[Maya Wang (mayawang)]

LGTM. The ParseInt change is in the diff at head now, so dberkov's point is covered.

atomic.Uint64 is worth more than tidiness here, it makes any future non-atomic access a compile error rather than a silent race, and it confirms line 92 is the only reader.

One optional nit, pre-existing, so ignore if you'd rather keep the diff tight. ParseInt with bitSize 64 accepts up to ~9.2e18, and time.Duration(d) * time.Second on line 81 overflows int64 nanoseconds above roughly 9.2e9 seconds. So ?duration=10000000000 wraps negative and time.Sleep returns straight away, which is the opposite of what an e2e test asserting graceful shutdown would expect. Atoi behaved the same way on 64-bit so it's not a regression, but an upper bound alongside the existing d < 0 check would close it.

[Maya Wang (mayawang)]

One more thing, spotted after I approved. The commit message still lists a change that isn't in the diff, same shape as Dima's point:

▎ Fix time.Tick leak: Replace time.Tick with time.NewTicker and ensure it is released via defer ticker.Stop()

Line 218 is unchanged, still for range time.Tick(10 * time.Second). The PR body only lists the two changes, so I think it's just the commit message that drifted.

I'd strip the bullet rather than implement it. That loop is the last statement in main() and never exits, so it's the case time.Tick's docs call fine, and a deferred Stop() wouldn't run anyway given the SIGTERM path goes through os.Exit.

Other two look good.