fix(acp): preserve filesystem unavailability - #7424

Summary

• classify ACP filesystem reverse-RPC timeouts and dropped response channels as KernelOpError::Unavailable
• make runtime file_read and file_write fall back to the local filesystem only for that unavailable class
• preserve hard failures for protocol and internal ACP errors
• add dispatch-level regressions for local read fallback, local write fallback, and non-fallback errors

Inventory finding: F-78fb8e62d385a9e7

Verification

• CARGO_TARGET_DIR=/private/tmp/librefang-target-pr7424-final cargo test -p librefang-acp
• CARGO_TARGET_DIR=/private/tmp/librefang-target-pr7424-final cargo test -p librefang-runtime acp_unavailable_file -- --nocapture
• CARGO_TARGET_DIR=/private/tmp/librefang-target-pr7424-final cargo test -p librefang-runtime acp_internal_file_errors_do_not_fall_back_to_local_filesystem -- --nocapture
• CARGO_TARGET_DIR=/private/tmp/librefang-target-pr7424-final cargo clippy -p librefang-acp -p librefang-runtime --all-targets -- -D warnings
• CARGO_TARGET_DIR=/private/tmp/librefang-target-pr7424-final cargo check --workspace --lib
• cargo fmt --all -- --check
• git diff --check origin/main...HEAD
• pre-commit and pre-push hooks

Out of scope

• changing ACP terminal reverse-RPC fallback behavior
• treating editor-returned protocol errors as local fallback conditions