feat(session): add EventCompaction type and compaction filtering in contents_processor

[QuentinBisson]

Ports the compaction persistence and content-assembly model from adk-python to adk-go. Fixes #1001.

What's missing today

EventActions has no Compaction field and buildContentsDefault has no filtering step for compaction markers. This means raw events are never replaced by their summaries when building the LLM context — compaction runs but has no visible effect on what gets sent to the model.

Changes

session/session.go

type EventCompaction struct {
    StartTimestamp   time.Time
    EndTimestamp     time.Time
    CompactedContent *genai.Content
}

Added Compaction *EventCompaction to EventActions. The summary lives here, not in Event.Content, so compaction markers are excluded from normal content rendering and distinguishable from real model output.

internal/llminternal/contents_processor.go

New applyCompaction pass called inside buildContentsDefault, after transcription aggregation and before the async-function-response rearrangement step:

1. Collect all compaction marker events (those with Actions.Compaction != nil).
2. Remove subsumed markers — if marker C2's range fully contains marker C's range, C is dropped (a later compaction already covers it).
3. Sort active markers by StartTimestamp.
4. Walk the event list: events whose timestamp falls within an active marker's range are replaced by a single synthetic event carrying CompactedContent; compaction marker events themselves are always removed from the output.

This mirrors _apply_compaction in adk-python's flows/llm_flows/contents.py.

Downstream reference

kagent PR kagent-dev/kagent#2025 uses this via a replace directive and exercises both sliding-window and token-threshold compaction modes against Anthropic and OpenAI adapters. The downstream compaction driver (watermark-based invocation counting, longestSelfContainedPrefix safety, summarizer LLM) lives in kagent — out of scope for this PR, which only adds the session primitives and content-assembly filtering that the driver depends on.

[google-cla]

Thanks for your pull request! It looks like this may be your first contribution to a Google open source project. Before we can look at your pull request, you'll need to sign a Contributor License Agreement (CLA).

View this failed invocation of the CLA check for more information.

For the most up to date status, view the checks section at the bottom of the pull request.