Add OpenCode agent support

[etothexipi]

Adds support for the OpenCode agent (https://opencode.ai), providing users with another powerful AI code review option. Enhances #2

Changes

• Add opencode agent implementation using non-interactive CLI mode
• Update fallback chain: codex → claude-code → gemini → copilot → opencode
• Update tests, README, and CLI help text
• Follows same pattern as gemini/copilot PR (Add gemini-cli, copilot CLI support #3)

Installation

npm install -g opencode-ai

Testing

• go test ./... passes
• go build ./... succeeds
• Follows established contribution patterns

Notes

OpenCode is an open-source AI coding agent that supports multiple backends, including local LLMs.

[etothexipi]

Tested with configs:

• .roborev.toml

agent = "opencode"    # codex, claude-code, gemini, copilot, or opencode
review_context_count = 5

# Project-specific review guidelines (multi-line string)
review_guidelines = """
We are not doing database migrations because there are no production databases yet.
Prefer composition over inheritance.
All public APIs must have documentation comments.
"""

• config.toml

server_addr = "127.0.0.1:7373"
max_workers = 4
review_context_count = 3
default_agent = "opencode"
codex_cmd = "codex"
claude_code_cmd = "claude"

Test review output:

Review: 768150d (opencode)
                          


## Review: Commit 768150d

**No issues found.**

I reviewed this commit for bugs, security issues, testing gaps, regressions, and code quality
concerns.

### Summary

This commit refactors the OpenCode agent's review command to use the new `opencode run` headless
invocation instead of the previous flag-based approach. Key changes:

1. **Command structure**: Switched from `opencode -c <path> -p <prompt> -q` to `opencode run
--format default <prompt>`
2. **Working directory**: Now sets `cmd.Dir = repoPath` instead of passing `-c` flag
3. **Error handling**: Improved to include both stdout and stderr in error messages

The changes are well-documented with clear comments explaining the new approach and helpful
references. The error handling improvement is a good enhancement that will make debugging easier. No
security, testing, or regression concerns identified.

[wesm]

@etothexipi thanks!!