…per provider

The /model picker currently renders one row per ``custom_providers``
entry. When several entries share the same provider name (e.g. four
``ollama-cloud`` entries for ``qwen3-coder``, ``glm-5.1``, ``kimi-k2``,
``minimax-m2.7``), users see four separate "Ollama Cloud" rows in the
picker, which is confusing UX — there is only one Ollama Cloud
provider, so there should be one row containing four models.

This PR groups ``custom_providers`` entries that share the same provider
name into a single picker row while keeping entries with distinct names
as separate rows. So:

* Four entries named ``Ollama Cloud`` → one "Ollama Cloud" row with
  four models inside.
* One entry named ``Ollama Cloud`` and one named ``Moonshot`` → two
  separate rows, one model each.

Implementation
--------------
Replaces the single-pass loop in ``list_authenticated_providers()`` with
a two-pass approach:

1. First pass: build an ``OrderedDict`` keyed by ``custom_provider_slug(name)``,
   accumulating ``models`` per group while preserving discovery order.
2. Second pass: iterate the groups and append one result row per group,
   skipping any slug that already appeared in an earlier provider source
   (the existing ``seen_slugs`` guard).

Insertion order is preserved via ``OrderedDict``, so providers and
their models still appear in the order the user listed them in
``custom_providers``. No new dependencies.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

… grouping

Tests for salvaged PRs #9233 and #8011.