Review the architecture split

[jlamypoirier]

🎯 Goal (What & Why)

Base model configs are currently split into architecture/non-architecture. This is a legacy behaviour that was introduced as a stopgap solution for:

1. Loading pretrained configs while allowing changes in parameters that support it.
2. Marking a subset of the parameters as essential to a model, which absolutely need to be kept fixed when loading a pretrained model (i.e. changing it would cause a crash or otherwise break the model).

(1) Is made obsolete by #211, and will become irrelevant after #231. (2) On the other hand needs a replacement. Ignoring the problem is not recommended because it would make the loading of pretrained checkpoints unsafe, causing cryptic crashes and unexpected breakdown. My suggestion (which I've been planning for a while) would be to replace the hard class split by a simple field descriptor, i.e. something like field hints. This would address the issue with relatively low added complexity (and could be made opt-in, ex. for experimental models).

🚀 Execution Plan

(This section may start as an incomplete draft but must be defined before implementation begins.)

Step 1: What is the smallest working version?

(Describe the simplest way to implement this feature with minimal effort.)

Step 2: What additional optimizations are possible (but optional)?

(List potential refinements that can be added in later PRs if needed.)

📌 Acceptance Criteria (Must-Haves for Completion)

• The feature must be functional and tested.
• The implementation must be documented in practical terms.
• The PR must include a performance/impact summary.
• No refactors unless directly necessary for feature completion.

🛠️ Project Management

• Assign the project to the Fast-LLM project.
• Set the Estimate field (in days) in the GitHub project.
• Use the Size field to categorize the PR size (Small/Medium/Large).
• Assign an owner when opening the issue.