feat(config): add global reasoning effort configuration with CLI, env, and shell-plugin support

[tusharmath]

Summary

Add global reasoning configuration support, allowing users to set and persist a reasoning effort level (none/minimal/low/medium/high/xhigh) that applies to all agents, with agent-level settings taking priority.

Context

Previously, reasoning effort could only be configured per-agent or via ad-hoc environment variables, making it cumbersome to set a consistent reasoning level across an entire session or globally. This PR introduces a first-class [reasoning] config block in ~/.forge.toml, a CLI config get/set reasoning command, and a shell-plugin :reasoning / :re shortcut so users can quickly switch effort levels interactively.

Changes

• New Effort enum (none, minimal, low, medium, high, xhigh) added to forge_config and exposed from forge_domain
• ReasoningConfig struct added to forge_config with enabled and effort fields, serialisable to/from TOML/JSON
• ForgeConfig gains a top-level reasoning field; schema (forge.schema.json) updated accordingly
• Provider DTOs updated — Anthropic, Google, and OpenAI reasoning transformers now map the new Effort variants to their respective native representations (token budgets / level enums)
• AppConfigService / API trait extended with get_reasoning_config / set_reasoning_config methods
• forge_infra env layer reads FORGE_REASONING__EFFORT to allow per-invocation overrides
• CLI — new config get reasoning and config set reasoning <effort> subcommands; built-in commands JSON updated
• Shell plugin — new :reasoning (:re) session action and :config-reasoning (:cr) persistent-config action backed by an fzf picker; _forge_exec / _forge_exec_interactive export FORGE_REASONING__EFFORT when a session override is active
• Dependency downgrades — google-cloud-auth, insta, uuid, config crate versions rolled back to resolve lock-file conflicts

Key Implementation Details

The effort level flows through three priority layers, from highest to lowest:

1. Agent-level reasoning block in the workflow YAML
2. Session override set via :re (stored in _FORGE_SESSION_REASONING, exported as FORGE_REASONING__EFFORT)
3. Global [reasoning] block in ~/.forge.toml

Effort is mapped to provider-specific tokens/levels:

Effort	Anthropic budget_tokens	Google Level	OpenAI effort
none	disabled	disabled	none
minimal	1 024	Minimal	minimal
low	2 048	Low	low
medium	8 192	Medium	medium
high	16 384	High	high
xhigh	32 768	High	xhigh

Use Cases

• Set effort = "high" globally for a deep-dive debugging session, then drop back to "medium" for routine work via :cr
• Use :re to override reasoning effort for just the current shell session without touching the persistent config
• CI/scripting: pass FORGE_REASONING__EFFORT=none to completely disable reasoning tokens and reduce cost

Testing

# Unit / integration tests
cargo insta test --accept

# Verify CLI commands
forge config get reasoning
forge config set reasoning high
forge config set reasoning none

# Shell plugin (in a zsh session with the plugin loaded)
:re        # interactive fzf picker — sets session-scoped effort
:cr        # interactive fzf picker — writes to ~/.forge.toml

Links

• Related PR series: reasoning effort groundwork in earlier PRs