fix: fall back to native text handling when FCP routing drops session_id

`slipstream read` failed with "session.open did not return session_id"
when the file's extension triggered FCP routing (e.g. .py with fcp-python
installed). The FCP passthrough response lacked session_id.

Now mirrors exec's existing fallback: retry with force_native=true when
session_id is missing, bypassing FCP to use native text handling.

fix: fallback to force_native when FCP session.open fails, v0.5.18

When fcp-python/fcp-rust fails to start (Docker, CI), session.open returns
an error without session_id. Now retries with force_native=true to fall
back to text mode. FCP still handles reads when working — emergent LSP
intelligence preserved. Only falls back on failure.

fix: always force_native in CLI exec — FCP passthrough breaks Docker/…

…CI, v0.5.17

fix: force_native on session.open when ops present, bypass FCP routin…

…g, v0.5.16

BUG-007 v2: after skipping FCP passthrough short-circuit (v0.5.14),
session.open still returned FCP response without session_id. Fix:
- Added force_native field to SessionOpenParams
- CLI sets force_native=true when --ops is present
- handler.rs skips FCP routing when force_native is set
- Result: session.open returns native text session with session_id

fix: ServerInfo instructions lead with batch ops as default pattern, …

…v0.5.15

Batch reads+edits in one ss(ops=[...]) call is now the first example,
not buried as an advanced option. Single-file operations are "shorthand."
This fixes LLM behavior of defaulting to single-file ss_session("read ...")
instead of batching — observed across multiple sessions.

fix: exec --ops no longer silently dropped on FCP passthrough, bump t…

…o v0.5.14

BUG-007: When a file was routed to an FCP handler (e.g. fcp-python for .py),
run_exec() and exec_one_shot() returned early without applying --ops edits.
Now only short-circuits on FCP passthrough when no ops are requested.
With ops, falls through to text-mode handling so str_replace works.

0% success rate across 12 attempts in SWE-bench v14 → should now work.

fix: 30s timeout on client request to prevent indefinite hangs, bump …

…to v0.5.13

Client::request() had no timeout on next_line().await — if the daemon
stopped responding (e.g. FCP handler hang), the client blocked forever.
Wraps the read in tokio::time::timeout(30s). Includes regression test.

v0.5.12: add batch read to ServerInfo instructions

v0.5.11: Fix CWD-relative path resolution

v0.5.10: Bug fixes for first-use experience

- Fix exec --ops crash on FCP passthrough responses
- Remove sed -i interstitial (always falls through to real sed)
- Skip interstitial when stdin is piped (no broken pipe in pipelines)
- Handle BrokenPipe silently in shim error handler
- Simplify verbose nested error messages (e.g. 'file not found')