Add per-session/per-partner thread-history window control (backend hook not yet available)

[zoharbabin]

Why

Issue #43 documents that thread history has no default size cap: the full transcript is sent as context on every turn, so per-turn token cost grows indefinitely on long-lived threads. Backend investigation for that issue confirmed a token-budget-based trimming/summarization mechanism exists server-side, but it is a global on/off switch today, with no per-partner or per-session control exposed.

Per that issue's own success criteria, any future SDK-level control for this must be tracked as a separate, explicit enhancement rather than silently assumed to exist. This issue is that tracking item.

What

Once the backend exposes a concrete hook (e.g. a per-partner or per-session parameter for the trimming/summarization budget, or a max-history-window setting), add a corresponding SDK-level control under ./management (Conversations or threads) so integrators can tune thread-history cost/behavior per session or per partner, instead of relying on the current global on/off switch.

Non-goals right now

• No SDK code should be written until a concrete backend hook exists to wrap. There is nothing to wrap today.
• This is not a request to change backend behavior; that decision belongs to the backend team.

Acceptance criteria

• A documented, testable SDK method/option exists for bounding or tuning thread-history size or cost on a per-session or per-partner basis.
• API-REFERENCE.md's Threads section is updated to describe the new control once it ships, replacing the "no control exists" disclosure.

Related

Related to the thread-history cost/growth documentation in API-REFERENCE.md (Threads section), which flagged this gap as a doc-only, backend-blocked issue at the time of writing.