version 0.3.5 of llama-cpp-python has a known issue abetlen/llama-cpp-python#1861 version 0.3.2 has granite 3.0 support and does not have this issue. Bump to this version

this required some additions to how we handle chat exceptions. As of these newer 0.3.z llama-cpp-python versions,
a bad request causes the server to die. This requires us to know the max_ctx_size of the server before passing a completions request so that
we can maintain the behavior of trimming messages until we can respond to one that fits.

in order to do this, the config now contains a `current_max_ctx_size` field that we will update when spinning up a server.
in the case that a user implicitly starts a llama-cpp-python server when calling `ilab model chat`, we set the max_tokens to the
current `max_ctx_size` in the serve config.









**Checklist:**

- [ ] **Commit Message Formatting**: Commit titles and messages follow guidelines in the
  [conventional commits](https://www.conventionalcommits.org/en/v1.0.0/#summary).
- [ ] [Changelog](https://github.com/instructlab/instructlab/blob/main/CHANGELOG.md) updated with breaking and/or notable changes for the next minor release.
- [ ] Documentation has been updated, if necessary.
- [ ] Unit tests have been added, if necessary.
- [ ] Functional tests have been added, if necessary.
- [ ] E2E Workflow tests have been added, if necessary.
<hr>This is an automatic backport of pull request #2825 done by [Mergify](https://mergify.com).


Approved-by: cdoern

Approved-by: alinaryan