[FEATURE] Support block/proc for dynamic context setup in Agent

Scope check

• This is core LLM communication (not application logic)
• This benefits most users (not just my use case)
• This can't be solved in application code with current RubyLLM
• I read the Contributing Guide

Due diligence

• I searched existing issues
• I checked the documentation

What problem does this solve?

RubyLLM::Agent has helpful macros for defining chat properties used (model, params, etc). There is a macro for context, but unlike the other macros, it only handles static values, and not procs/blocks evaluated at runtime. For cases where units of work are defined as subclasses of RubyLLM::Agent, there's no easy way to inject instance-specific details (eg: retry limits, timeout limits, per-tenant API keys).

Proposed solution

Following the conventions as the other macros, make the context macro take either a static value or a proc, which can read values from input and other sources at runtime. Happy to provide a PR if it helps.

Why this belongs in RubyLLM

Making the context macro follow the same value-or-proc-based pattern as the other macros makes this work without having disassemble the Agent into its constituent parts in applications (which would be a shame, it's a nicely designed class).

Start at RubyLLM::Agent's context macro and compare it with the other chat-property macros described in the issue. Verify the existing Agent behavior before changing it; done means context supports both static values and runtime procs or blocks, including instance-specific input.