[fix][fs] Fix warmup query hanging for 10 minutes after warmup has finished

[Wine93]

Problem

dingo fs warmup add <path> (without --daemon) leaves the progress bar stuck at 100% for exactly 10 minutes before the command returns.

Observed on a live cluster: the warmup itself finished in ~20ms, but the CLI kept polling for 10 more minutes.

16:08:16  dingo fs warmup add ... starts
16:08:18  client: Set warmup task context: [key: 20000000001, ...]
16:08:18  all 5 blocks fetched, task complete
16:08:19  client: Get warmup task status value: 5/5/0   <- already done at the first poll
   ...    4938 polls, every 200ms, value never changes
16:18:16  process finally exits                          <- exactly 10 minutes later

Root cause

The two sides disagree on what "finished" means.

Client — WarmupManager::FinishTask keeps a completed task's snapshot in finished_status_ for kFinishedStatusTtl = 10 minutes, so GetWarmupTaskStatus keeps returning 5/5/0 for that whole window and only returns 0/0/0 once the TTL expires.

CLI — the poll loop in runQuery treated total == 0 as the only completion signal, so it had to wait out the client's TTL.

This is a regression from dingofs 8e728fdd7 ("[feat][client] Make warmup scheduling nonblocking"). Before that commit a finished task was dropped from warmup_tasks_ immediately and the status went to 0/0/0 right away, which is what the total == 0 check implicitly relied on. Retaining the finished snapshot is intentional on the client side (otherwise warmup query cannot distinguish "not started" from "just finished"), so the fix belongs here.

Fix

Break once every block is accounted for — finished + errors >= total — which is the actual completion signal. The total == 0 case is kept for when the client has already dropped the status. bar.Set64 runs before the new break so the bar ends at 100% rather than at a partial value.

Also corrects a comment and an error message that described the payload as [finished/total/errors]; the client emits [total/finished/errors] and the parsing already used that order.

Testing

• go build and go vet pass for cli/command/fs/warmup/; gofmt clean.
• Not verified end-to-end. go build ./... fails in this repo, but identically on a clean main (58 undefined: errors in proto/dingofs/proto/{cache,web} and internal/rpc/mds.go — the generated proto code is out of sync with the Go code). I could not produce a dingo binary to re-run the scenario against a live mount. Please build in a working environment and confirm dingo fs warmup add now returns in seconds instead of 10 minutes before merging.