Telegram: TestCoreExpiredTap is flaky #51

Why

TestCoreExpiredTap in internal/telegram failed once in a full go test -race ./... run on 2026-10-15 ("0 reports, want 1"). It passed 300 of 300 runs alone and in every later full run.

The likely cause: poll.go answers the tap (answerCallbackQuery) before it reports the refused tap, and the test checks the reports right after it sees the answer, without waiting for the report. A flaky test in CI would block merges and teach people to re-run red builds.

Also flaky: TestCoreRedeliveredTapKeepsResult (attack_test.go:202: "waited 10s in vain for 1 calls of editMessageReplyMarkup"). It failed in CI on PR #52 (2026-10-15), and on main 03bb431 it failed 29 of 120 runs with go test -race -count=40 -cpu 1,2,4 -run TestCoreRedeliveredTapKeepsResult ./internal/telegram/.

What

• The test waits for the report (with a deadline), or the code reports the tap before it answers it, whichever matches DESIGN.md section 7.
• Look for the same pattern in the other TestCore* tests.

Done when

• go test -race -count=200 -cpu 1,2,4 -run 'TestCore' ./internal/telegram/ passes.
• No test in the package checks a report without waiting for it.