[CI] Workload package downloads time out or reset during installation

## Failure signature

`dotnet workload install` fails after Azure Artifacts package downloads time out or reset:

```text
Failed to download package '...'
The download ... timed out because no data was received for 60000ms.
Unable to read data from the transport connection: Connection reset by peer.
Workload installation failed. Rolling back installed packs...
```

Stable signature: `workload-package-download-timeout-or-connection-reset`

## Confirmed occurrences

| PR / head | Azure run, attempt, job, and task | Worker / terminal time (UTC) |
|---|---|---|
| #26548 / `40484470929660b603770c71c1402bb0b2756ada` | [15194958 attempt 1 / Detect API changes / Build](https://devdiv.visualstudio.com/0bdbc590-a062-4c3f-b0f6-9383f67865ee/_build/results?buildId=15194958&view=logs&j=276491e5-a85e-51e3-6c38-98e261218d1a&t=76afb9de-7feb-5cd3-1173-f84cb7d12f5e) | `AcesShared 15`, 2026-09-03 05:01:16 |
| #26552 / `c7127604c986af36f8f97616117e03761dc5cc19` | [15196658 attempt 1 / `T: cecil` / Install dotnet workloads](https://devdiv.visualstudio.com/0bdbc590-a062-4c3f-b0f6-9383f67865ee/_build/results?buildId=15196658&view=logs&j=d43b0c36-8a1f-5afc-8ffd-00758b82d008&t=15247d70-e5dc-5287-0021-f725d00d33ca) | `AcesShared 15`, 2026-09-03 08:30:18 |

**Total:** 2 confirmed root-task occurrences across 2 distinct monitored PRs.

First observed: `2026-09-03T05:01:16Z`. Latest observed: `2026-09-03T08:30:18Z`.

## Classification and impact

**CI infrastructure failure.** The same transport failure occurred on unrelated PRs and in different jobs, but on the same worker. The requested packages were located and some retry downloads succeeded; installation ultimately failed because other downloads timed out or the connection reset. This is distinct from a missing-package/source-configuration failure.

The affected jobs fail during workload setup before product build or test execution. #26548's API-diff prerequisite failed before any current-PR or stable-baseline comparison, so its subsequent result-publication failure was only a missing-artifact cascade and does not authorize a binding or suppression-file change. #26552's missing TestSummary/HtmlReport messages are likewise downstream cascades.

## Investigation performed

- Inspected authenticated Azure build metadata, timelines, root task logs, workers, attempts, and package-download diagnostics.
- Correlated the 60-second no-data timeouts, `Connection reset by peer`, workload rollback, and common `AcesShared 15` worker across both PRs.
- Searched all open and closed `dotnet/macios` issues separately by the normalized signature, timeout text, package-download text, NuGet Manager component, and workload-install task. No matching issue was found.

Azure logs and artifacts are retention-limited.

## Retry status and workaround

- #26548: Azure accepted a failed-jobs retry for API-diff stage `generate_api_diff`. Attempt 2 succeeded on `AcesShared 6`; the workload-download failure did not recur. Retry usage is **1 lifetime / 1 head**.
- #26552: build 15196658 remains active, so no retry has been submitted. Its complete failure set and retry eligibility must be reconciled after the parent build becomes terminal.

A failed-jobs retry may clear an isolated occurrence after refreshing the PR head, mergeability, retry budgets, and equivalent active attempts. The infrastructure fix is to address the package-download transport reliability affecting the worker/feed path.

Start with the authenticated Azure build metadata, timelines, root task logs, worker details, and package-download diagnostics described in the issue. Confirm the timeout and connection-reset pattern on the worker/feed path; done means workload package downloads complete reliably without the installation rollback or downstream CI failures.

Written by the indexing model from the issue text.