fix(core): read OpenCode configs saved with a UTF-8 BOM - #1510

Problem

parse_jsonc hands the file straight to serde_json, which rejects a leading UTF-8 BOM. An OpenCode config saved with a BOM (Windows editors still do this) fails to parse, and the JSONC fallback doesn't strip the BOM either. Every configured model name in that file is then silently ignored, and reports fall back to the raw model ids.

OpenCode itself reads such a config without complaint, because its file read decodes the text as UTF-8, which drops the BOM. The same parse_jsonc is also used for omo.json / omo.jsonc in scanner.rs, so an OmO config with a BOM loses its task.state_dir the same way.

Fix

parse_jsonc strips one leading U+FEFF before parsing, for both the strict JSON and the JSONC path.

Tests

New reads_config_saved_with_a_utf8_bom covers plain JSON and JSONC (comment plus trailing commas), both BOM-prefixed.

I couldn't build the whole crate on my machine: vendored OpenSSL needs make, which this box doesn't have. So I checked the change by running parse_jsonc in a scratch crate with only serde_json. A BOM-prefixed document parses with the fix and fails without it. The new test went through cargo fmt but hasn't run inside this repo; CI will run it.

AI-assisted: I directed the investigation and reviewed and verified the change myself.

Summary by cubic

Fixes parse_jsonc so it strips a leading UTF-8 BOM before parsing. OpenCode and omo.json/omo.jsonc configs saved with a BOM (common from Windows editors) previously failed to parse, silently losing every configured model name (reports fell back to raw model ids) and task.state_dir.

Adds a regression test covering both the strict JSON and JSONC paths with a BOM prefix.

^{Written for commit 71b4d36. Summary will update on new commits.}