Teach both installers the telemetry opt-out's new name

[mattpodwysocki]

Ported from mapbox-cli-private#141, which can no longer merge there now that oss/ is a submodule (#132).

The gap

The binary's telemetry opt-out was renamed from DISABLE_TELEMETRY to MAPBOX_CLI_NO_TELEMETRY, and that break was documented honestly — the changelog says a DISABLE_TELEMETRY=1 "silently stops opting out". Both installers were left on the old name, so after that release neither variable silenced both halves:

Set	mapbox binary	install.sh / install.ps1
MAPBOX_CLI_NO_TELEMETRY=1	silent	still reports
DISABLE_TELEMETRY=1	reports (documented)	silent

The private repo's docs/distribution.md already claimed the fix:

MAPBOX_CLI_NO_TELEMETRY switches all of that off … Both installers carry their own copy of this logic, and both test suites pin it in both directions.

The first half was false, and the second pinned the wrong name. So this makes the documentation true rather than changing it.

The old name keeps working, in the installers only

Asymmetric on purpose. The binary's break was announced in a changelog someone upgrading can read. An installer is fetched and executed in one line — curl … | sh, irm … | iex — so its reader has no release notes in front of them, and breaking an opt-out is the one change that must not happen quietly. The comment in each file says exactly that, so the asymmetry reads as chosen rather than as an oversight in the other direction.

When both are set the new name wins, including when it declines: an explicit MAPBOX_CLI_NO_TELEMETRY=0 beats a stale DISABLE_TELEMETRY=1 left in an image from before the rename. Without that the old variable could never be retired.

Verified rather than reasoned about

I pulled each script's decision block out and ran it over the cases directly — nine in sh, eight in pwsh 7.6.6 — covering both names, both precedence directions, the six false spellings, an unknown spelling (which must send less), whitespace, and a set-but-empty new name as a deliberate clear. All pass.

Then added cases to both real harnesses, which pass here too:

MAPBOX_CLI_NO_TELEMETRY is honoured, and outranks the old name
  ok    exits 0 — the install is not what is being switched off
  ok    still names the installer
  ok    no platform rides behind it
  ok    and no source tag either
  ok    the new name wins when the two disagree
  ok    and wins in the opt-out direction too

Both suites green end to end in this repo: scripts/test-install.sh reports all cases passed, scripts/test-install.ps1 reports all good (2 pre-existing Windows-only skips, unrelated).

new_case_env / New-CaseEnv now clear the new variable too, for the same reason they already cleared the old one — a developer with it set in their own shell would otherwise turn every marker case into a failure that looks like the marker broke.

The original PR also edited docs/distribution.md, which lives in the private repo and stays there.

[zmofei]

LGTM — precedence logic and installer parity verified, tests cover both directions, CI green. Note: PR has a merge conflict with base, needs rebase before merging.