feat: allow specifying callback port for OIDC auth

Inizia in crates/openshell-cli/src/oidc_auth.rs intorno alle righe 188-190 e segui come la porta del listener arriva alla costruzione di redirect_uri. Aggiorna docs/reference/gateway-auth.mdx intorno alle righe 92-98 e 123, quindi aggiungi la copertura dei test unitari per la porta configurata e un errore di collisione del bind. Il lavoro è completo quando la variabile d’ambiente seleziona la porta del listener, le porte non valide o non disponibili producono errori utilizzabili e il comportamento effimero predefinito rimane invariato.

User Story

As an openshell consumer, I'd need the OIDC flow to work with a specific ephemeral port for SSO login. This is because the underlying auth0 provider does not allow regex on the port for allowed callback urls

Problem Statement

Provide a way to fix the CLI ephemeral callback port to a specific value so that it's easier to add the callback url

Impact / Why This Matters

Right now the SSO login requires whitelisting every possible ephemeral port in localhost which is not ideal.

Proposed Design

Provide the control of port via env variable OPENSHELL_OIDC_CALLBACK_PORT or something similar

Acceptance Criteria

• the callback port gets opened on specific port instead of any ephemeral port

Alternatives Considered

client credentials flow is what I am considering it for now

Agent Investigation

Spike Plan: feat: allow OIDC callback listener to bind a fixed port via env var
Scope decision: Env-var only (OPENSHELL_OIDC_CALLBACK_PORT) — no GatewayMetadata/clap changes.
Problem statement: The CLI's OIDC login flow binds an OS-assigned ephemeral port (127.0.0.1:0) for the local callback listener. Providers that require pre-registering exact redirect URIs can't reasonably whitelist the full ephemeral range, so login fails or requires impractical whitelisting.
Code changes (all in crates/openshell-cli/src/oidc_auth.rs):

• :188 — check OPENSHELL_OIDC_CALLBACK_PORT; if set, parse as u16 and bind that specific port with a wrapped, actionable error on bind failure; else keep current bind("127.0.0.1:0") behavior.
• No changes needed to redirect_uri construction (:189-190) or callback handling (:660-716) — they already work off whatever port the listener has.
  Labels: area:cli, state:validated
  Complexity: Low · Confidence: High · Est. files: 1 (+ docs, + tests)
  Risks to include in the issue:
• Bind-failure UX must give actionable guidance (today's raw OS error is not user-friendly).
• Fixed port is predictable, but existing CSRF state check + PKCE already prevent exploitation — mention, not a blocker.
• No interaction with token/mTLS storage or TUI (confirmed).
  Docs impact: docs/reference/gateway-auth.mdx — add the new env var to the table (~line 92-98) and update prose at line 123.
  Test additions: unit test setting the env var and asserting the bound port matches; a bind-collision test asserting the friendly error message.

Checklist

• I've reviewed existing issues and the architecture docs
• This is a design proposal, not a "please build this" request