pageserver: eliminate 500 cases during timeline creation/deletion

[jcsp]

Problem

Some code paths munge cancellation errors into anyhow::Error and consequently surface it as a 500 in the HTTP API.

Impact

These 500s make it harder to alert on a produciton system and risk hiding real issues: 500s should be reserved for unexpected situations, whereas deleting a timeline on a tenant that's migrating (shutting down) is expected during migrations (e.g. during a deploy)

In test_timeline_archival_chaos we have allow lists like this:

    env.storage_controller.allowed_errors.extend(
        [
            ".*error sending request.*",
            # FIXME: the pageserver should not return 499s on cancellation
            ".*InternalServerError(Error deleting timeline .* on .* on .*: pageserver API: error: Cancelled",
        ]
    )

    for ps in env.pageservers:
        # We will do unclean restarts, which results in these messages when cleaning up files
        ps.allowed_errors.extend(
            [
                ".*removing local file.*because it has unexpected length.*",
                ".*__temp.*",
                # FIXME: there are still anyhow::Error paths in timeline creation/deletion which
                # generate 499 results when called during shutdown
                ".*InternalServerError.*",
                # FIXME: there are still anyhow::Error paths in timeline deletion that generate
                # log lines at error severity
                ".*delete_timeline.*Error",
            ]
        )