[wu-sheng]

Read a tool call as its arguments in AI agent conversations

• If this is non-trivial feature, paste the links/URLs to the design doc.
• Update the documentation to include this new feature.
• Tests(including UT, IT, E2E) are added to verify the new feature.
• If it's UI related, attach the screenshots below.

This follows apache/skywalking-ai-sessionizer#94, which changes the text of a step that stands on one part of a record.
The OAP mirrors the document the Sessionizer defines in
The asz.view document.

• The rule. A step on a part that holds data reads that part alone: the prompt inside it when it is a
  queued_command envelope, and otherwise the data as the part holds it. A tool call that shares a record with the
  agent's text reads as its arguments, never as that text.
• Before. ConversationViewBuilder.fill took the readable text of the whole record whenever it differed from the
  part's data. So a tool call beside the agent's text read as that text, and a queued command beside other text read as
  that other text.
• How often. On a local Sessionizer root, 10 of 25 LangChain tool calls shared a record with text, and 3 of 109,397
  Claude Code ones.

Tests.

• RecordReadingTest.aStepOnAPartReadsThatPart mirrors the Sessionizer's own test: a tool call beside the agent's text
  reads as its arguments, and a queued command beside unrelated text reads as its own prompt. fill is now
  package-private so the test can call it.
• Each half was proven by reverting it and seeing the test fail: with the old rule the call read as the agent's text,
  and with the prompt taken from the whole record the queued command read as the unrelated text.
• No fixture holds a record with text beside a call or a queued command, so every document the tests compare with the
  Sessionizer's is unchanged. All 186 tests of the module pass.

The Horizon UI needs no change: it draws the text the document carries.

• Update the CHANGES log.