Update architecture spec to match current implementation

[lore-agent]

The HAL Engine architecture specification has drifted significantly from the current implementation (100% divergence).

• Review specs/hal-engine-architecture/spec.md against actual codebase
• Document current architecture decisions and any intentional deviations
• Update spec to reflect reality or align implementation with spec intent

This keeps our documentation trustworthy for new contributors and future maintenance.

Missing symbols (heuristic)

• class: ThinkingTagParser — Parser that separates thinking tags from user-facing text during streaming and handles partial tags at chunk boundaries
• class: ThinkingEntry — Data structure for collapsible thinking segments in the UI
• class: AssistantEntry — Data structure for text segments rendered as markdown

What you should actually do

• Decide the direction first: is the spec stale, or is the code wrong? For a
  reconstruction-grade spec the answer is almost always "update the spec".
• If you update the spec, fix the diverged items above and re-verify every
  ([validated by …](…)) link and #Lnn anchor on the statements you touch.
• If this is a false positive — the named items are endpoints, fields, or methods
  rather than top-level symbols, and the behaviour still matches — close this as
  stale rather than editing the spec.

---

Managed by Lore · created by spec-drift · Lore-Task: 86f299b9-6d79-485c-8895-b0cf0908baf1

[lore-agent]

Agent lore-agent-86f299b9 picked up this task. Follow it on the pipeline: https://lore.gcp.re-cinq.com/tasks/86f299b9-6d79-485c-8895-b0cf0908baf1