## Context

`Invoke-GitlabApi` pages by handing `Invoke-RestMethod` the
`-FollowRelLink` flag and letting it walk the `rel=next` chain GitLab
puts in its `Link` header. PowerShell 7.6 drops the `Authorization`
header when it follows one of those links, so the request for page 2
goes out unauthenticated. Anything using `-All` over a result longer
than one page breaks — GitLab serves 20 per page by default, so that is
most of them. `-PreserveAuthorizationOnRedirect` keeps the header
attached.

The upstream bug is
[PowerShell#27861](PowerShell/PowerShell#27861);
the properly scoped fix is in flight as
[PowerShell#27981](PowerShell/PowerShell#27981)
(with [#27873](PowerShell/PowerShell#27873) as
the forward-port). This comes back out once that ships — hence the
comment pointing at the issue rather than a bare flag.

## Review guide

**Core change, and the whole diff** —
[`src/GitlabCli/Utilities.psm1`](https://github.com/chris-peterson/pwsh-gitlab/pull/167/changes#diff-bcad0c863b627b9e28cb755bd24ac1be9ab715a01e6750820f02009a80ec2c3cR103)

## Approach & trade-offs

The flag is broader than what this deserves: it also preserves the
header across *cross-origin* redirects, which is the case the strip was
added to close. Every `rel=next` GitLab emits here points back at the
same host the request already went to — measured, not assumed — so the
exposure it reopens is theoretical while the breakage it repairs is not.

The narrower alternative was switching the header to `PRIVATE-TOKEN`,
which PowerShell doesn't strip. It works, but it's GitLab-specific and
swapping the auth scheme to dodge a client bug is a bigger change to
reason about than one flag with an expiry date on it.

## Validation

Measured against `gitlab.getty.cloud` with a raw `Invoke-RestMethod`, so
the result isn't shaped by this module. Both endpoints page at 20 per
request:

| Endpoint | without the flag | with the flag |
| --- | --- | --- |
| `issues?assignee_username=…` (112 items) | **401 on page 2** | 112
items |
| `merge_requests?…&scope=all` (55 items) | **21 items, exit 0, no
error** | 55 items |

The second row is the one worth pausing on. The same defect does not
always surface as a failure — that call returned a partial result and
reported success, and a short list looks like an answer. This module's
callers have no way to tell 21 from 55.

The effect is endpoint-specific rather than uniform: on GitHub's API the
plain `/issues` list endpoint throws the same 401 while `/search/issues`
walks its chain intact. So which symptom you meet, or whether you meet
one at all, depends on the call you reach for first.


[`pwsh-github#11`](chris-peterson/pwsh-github#11)
carries the identical one-line fix. The two are independent — neither
has to merge first.