tui: friendlier model display + group same-endpoint providers in picker

[antydizajn]

What

Two related TUI quality-of-life fixes for users running multiple models behind a single proxy/aggregator endpoint (e.g. Palantir Foundry, Bedrock, self-hosted vLLM behind a single key).

1. Friendlier active-model name in the status bar

Long catalog IDs (Palantir RIDs like ri.language-model-service..language-model.anthropic-claude-4-7-opus) were truncated to ri.language-model-ser… by the existing 26-char slash-split path in _get_status_bar_snapshot(), leaving no useful identity in the status bar.

The status bar now:

• Reverse-looks up the active model id in model_aliases: / model.aliases: and shows the shortest configured alias when one exists. So a user who set opus47 as an alias gets opus47 in the status bar for free.
• Falls back to stripping Palantir's ri.<service>..<ns>. RID prefix before length-truncation, so the truncated label carries the actual model identity (anthropic-claude-4-7-opus) instead of the URN scheme.
• Reverse-alias map is cached at module level (config is loaded once per session; no need to re-resolve on every status-bar refresh).

2. Group same-endpoint providers: entries in the model picker

list_authenticated_providers() section 3 now groups providers: entries by (api_url, key_env, api_mode), mirroring section 4's existing grouping for custom_providers: lists.

Before: a Palantir Foundry config with two Anthropic-proxy entries (palantir-claude46 + palantir-claude47) produced two near-duplicate picker rows labelled "Palantir Claude 4.6 Opus" and "Palantir Claude 4.7 Opus" — same endpoint, same PALANTIR_TOKEN, same anthropic_messages wire protocol, differing only by model id.

After: those entries collapse into a single "Palantir Claude" row with both models in the dropdown.

Same-host entries with a different api_mode (e.g. an OpenAI-compat palantir-gpt54 alongside the Anthropic claude rows on the same host) keep distinct rows since the wire protocol differs — the same safety invariant section 4 already enforced for custom_providers:.

Group display name strips per-version trailing tokens ("Palantir Claude 4.7 Opus" → "Palantir Claude") only when the prefix has ≥2 words, so single-word names aren't over-trimmed.

The new code records (raw_display_name, api_url) into _section3_emitted_pairs for every raw entry that joined a group, so section 4's compatibility-merged custom_providers view (built by get_compatible_custom_providers() calling providers_dict_to_custom_providers()) still dedupes against the grouped row.

Why

Hermes already groups custom_providers: entries by endpoint+credential+protocol (section 4). The same shape exists in providers: (the keyed schema), but section 3 emitted one row per key, so any user with multiple models behind a single proxy hit a polluted picker. This brings sections 3 and 4 to parity.

The status-bar truncation made the long-RID problem worse because the shown prefix carries no per-model identity at all.

Manual smoke test

Local config with three Palantir entries (palantir-gpt54, palantir-claude46, palantir-claude47):

• Before: 3 picker rows.
• After: 2 picker rows (1 row "Palantir Claude" with 2 models in the dropdown, 1 row "Palantir GPT-5.4" with 1 model — separate because api_mode differs).

Status bar with provider=palantir-claude47, model=ri.language-model-service..language-model.anthropic-claude-4-7-opus:

• Before: ⚕ ri.language-model-ser…
• After: ⚕ anthropic-claude-4-7-opus (or any user-configured short alias if model_aliases: is set).

No behaviour change for users with the simple case

Single-entry providers: rows (one model per provider key) keep their existing label and slug — they're a single-member group and the per-version-token strip heuristic only kicks in when the prefix has ≥2 words.

[teknium1]

Thanks for addressing a real picker/status-bar usability gap. The current main behavior still has both premises: cli.py:4556-4560 truncates opaque IDs directly, and hermes_cli/model_switch.py:2017-2121 emits one section-3 row per configured provider.

Problems

• The proposed section-3 key at hermes_cli/model_switch.py:1593 omits extra_headers. Current section 4 deliberately includes normalized headers in its key (hermes_cli/model_switch.py:2222-2231), because same URL/key/mode entries can be distinct tenant-routed endpoints. This would collapse them and use the first entry's configuration for discovery.
• The manual list parsing at hermes_cli/model_switch.py:1600-1607 regresses current support for models: [{"id": ...}]. Main centralizes that contract in _declared_model_ids() (hermes_cli/model_switch.py:55-100) and section 3 uses it at :2046.

Suggested changes

• Salvage the grouping atop current main, include normalized headers in its identity, and use _declared_model_ids().
• Add focused section-3 regression tests for header-routed endpoints and list-of-dict model declarations.

Automated hermes-sweeper review.

[teknium1]

Please include normalized extra_headers in this identity. Current section 4 does so because same URL/key/mode entries can route to different tenants through headers; collapsing them selects the first entry's headers/configuration for the group.

[teknium1]

Please aggregate via _declared_model_ids(cfg_models) instead. This loop appends dicts for supported models: [{"id": "..."}] declarations, while current main preserves their string IDs through the shared parser.

[teknium1]

Merged via PR #67908 — both of your commits were cherry-picked onto current main with your authorship preserved in git log. Two follow-ups were applied on top per the earlier review: the section-3 group key now includes normalized extra_headers (matching section 4's invariant for header-routed tenants), and model declarations go through _declared_model_ids() so models: [{id: ...}] rows keep working. The gateway model-switch handler had moved to gateway/slash_commands.py since you branched, so the display-form edits were re-applied there. 8 regression tests added covering the fold, api_mode/header separation, and RID stripping. Thanks for the thoughtful QoL work — the centralized format_model_for_display refactor in your second commit was exactly the right shape.