Archive: files skipped after routing are not removed from the final new-byte total

[woutervanranst]

Split out of #165 (review thread).

The problem

ArchiveCommandHandler Stage 3 adds each routed file's size to incrementalSize, then publishes the exact figure once routing drains:

await _mediator.Publish(new RoutingCompleteEvent(Interlocked.Read(ref incrementalSize)), cancellationToken);

JobSink.SetNewByteTotal stores that as _newByteTotal and flips _newByteTotalFinal, after which the ETA's upload denominator is treated as exact.

But a file can still be skipped after it was routed — the large-upload stage skips a file whose local read fails between hashing and upload (ArchiveCommandHandler.cs, the fs.OpenRead try/catch that publishes FileSkippedEvent). Those bytes stay in the denominator and are never credited to uploaded, so:

• uploadEta = (newByteTotal − uploaded) / transferRate keeps a nonzero residual for a job that has actually finished uploading;
• the final pct sits below 100 %.

Note on the ordering

RoutingCompleteEvent fires when routing drains, which is before uploads finish — so this cannot be reconciled at publish time. The sink has to learn about the skip afterwards.

Sketch

• Add long Size = 0 to FileSkippedEvent (defaulted, so the ~existing call sites are untouched).
• Pass upload.FileSize at the post-routing skip sites only. The hash-stage skip must keep Size = 0 — those files never reached incrementalSize.
• Add a forwarder that subtracts from the sink's _newByteTotal.
• Check whether the small-file/TAR upload path has an equivalent skip.

Severity

Low: needs a file to become unreadable in the seconds between hashing and upload. The visible effect is a small residual ETA and a sub-100 % pct at the end of a run — arguably honest, since that file genuinely was not archived.