fix(claude,web): surface assistant content lost to a mid-stream API retry

[Antisophy]

Problem

When the Anthropic API retries a request mid-stream, the Claude CLI can stop emitting stream_event partials for the rest of that run. cydo builds the visible message only from those content_block_* stream events, so when they stop, the assistant turn can end up rendering as nothing, leaving the reply only inside the session-complete result block, which itself shows collapsed as a bare checkmark divider, leaving discovery of the full response very unlikely. When it happened to me, the only way I discovered that I actually was receiving responses was through a desktop notification I was lucky to catch when cydo wasn't focused, leading me to click around to try to find the response I'd seen in the notification and finally expand the checkmark divider, which I'd previously dismissed as uninteractive.

Fix

Two commits, the cause and a safety net:

• fix(claude): promote assistant content when stream events are missing: when a complete assistant event arrives with no active stream blocks, promote its content blocks straight to item/started + item/completed (the same way the sub-agent path already does) so the turn renders. Promoted item ids use a session-wide counter (these per-block events carry no stream index), and an api_retry system event resets stream tracking so a half-opened block from the aborted attempt can't leave stale state that suppresses the fallback.
• fix(web): keep session result expanded when its text never rendered: the result event always carries a copy of the reply; when that copy is the only surviving record, compare it to the last main-turn assistant message and, if it isn't already on screen, start the result block expanded (like error results already do) instead of a bare checkmark.

[Antisophy]

Split the backend change into two commits so the behavior fix is reviewable on its own: fix(claude): promote assistant content... is just the fallback plus the api_retry reset, and the following test(claude): unit-test assistant content promotion extracts the per-block promotion into a pure static promoteContentBlocks and adds the unit test. (The web commit already carried its own vitest coverage.)

Reason for the extraction: the promotion lives on ClaudeCodeSession, whose constructor spawns the claude process, so translateAssistantLive can't be driven from a unit test directly. Moving the per-block loop into a pure function (and hoisting the ClaudeBlock parse struct to class scope) makes it testable: a stream-less text block must become item/started + item/completed, and a tool_use block must keep its own id. The extraction is behavior-preserving; translateAssistantLive just calls the helper and emits the results.