Fix --timeout help description

openai: do not force API key, fix error processing

OpenAI backends will no longer require an API key to be provided. This
is because users may wish to use a service such as Portkey with its
Virtual Keys feature, which doesn't require an API key. If an API
key is not provided, an Authorization header is not sent.

Also, error handling in the openai implementation is both fixed and
expanded to support Portkey error format.

Allow extra headers for specific conversations too

This commit adds support for setting extra HTTP headers for all
requests issued as part of a specific conversation. These headers
will be sent in addition to any extra headers defined for the
backend itself, if any, and will take precedence over them.

The bedrock implementation does not support extra headers at all
at this point.

Bugfix: main API doesn't expose previous messages cap

The previously added capability to add previously exchanged
messages was only exposed in the individual provider
implementations and not in the main libaiac API.

Allow conversations to have "previous messages"

This commit modifies the `Chat` method (in a backwards compatible
way) to allow users to start the conversation with previously
exchanged messages, i.e. messages that had already been exchanged
between the user and the assistant. This practically allows
"loading" previous conversations in order to continue them.

To support this, the `Chat` method now takes `types.Message`
values in a variadic way. The messages field of the `Conversation`
type is also exposed now, so that users may save it.

Fix wrong version major version in goreleaser.yml

Introduce config files, multiple backends, refactor

This commit introduces a major API change for `aiac` in both its
CLI and library forms, as per feedback from the community:

- `aiac` now uses a configuration file where various named backends
  may be defined, each targeting a specific LLM provider and
  environment. For example, one may configure a staging Amazon
  Bedrock environment, and a production environment. Users refer to
  these backends using the names they select for them. Command line
  flags such as `--api-key` and `--aws-profile` are no longer
  accepted and are now part of the configuration file.

- `aiac` no longer hardcodes supported models for each LLM provider.
  Users can choose any model they want, and set a default model for
  each backend. The `--list-models` flag (previously the `list-models`
  subcommand) contacts the LLM provider's API to retrieve a list of
  supported models.

- The `aiac` CLI is now simpler, and is no longer comprised of
  multiple subcommands, meaning flags such as `--model` no longer
  need to be provided before the "get" argument (which was actually
  a subcommand before), they will be accepted anywhere (outside of
  the actual prompt). The word "get" or "generate" no longer need to
  be used at all either.

- `aiac` no longer errors out when the provider API response notes
  that the result was truncated. Apparently, this isn't always
  accurate. Instead, the "stop reason" returned by the API is
  returned to the user as part of the output.

Refer to the [README.md](README.md) file for more information,
including how to upgrade from a previous version.

Resolves: #91, #121, #123

Fix release workflow

The release workflow stopped working. This commit fixes it and
also updates actions used by it. To simplify the project management,
the Homebrew formula is also moved from the "homebrew-aiac"
repository into this repository.

Add support for the Ollama backend

This commit introduces support for the ollama.ai backend. Ollama is
an open source LLAMA backend meant for local usage. It supports
many different models, many of them related to code generation.

Usage of the ollama backend is available via the `--backend` flag
or `AIAC_BACKEND` environment variables set to `"ollama"`. Ollama
doesn't support authentication currently, so the only related flag
is `--ollama-url` for the API server's URL, but if not used, the
default URL is used (http://localhost:11434/api). With this commit,
`aiac` will not yet support a scenario where the API server is
running behind an authenticating proxy.

Resolves: #77

Add ability to save and continue chatting

The aiac prompt line, after showing results from the LLM provider,
allows saving only once, as it exits immediately after saving. This
commit introduces the ability to save and continue chatting, thus
allowing users to save multiple files during a conversation with
the model.

For backwards compatibility purposes, the "s" key will still save
and exit, but the new "w" key can be used to save and continue
chatting.