refactor(orchestration): remove tool-calls.json and the prompt journal

[teriyakichild]

Removes two persistence surfaces that predate OTel support and now duplicate what tracing provides.

Fixes: #637
Refs: #636

tool-calls.json → in-memory tool traces

The per-attempt tool-calls.json files embedded every tool output in full, and append_tool_call re-read, re-parsed, and rewrote the whole file on each call while holding the shared persistence mutex. This was the fourth on-disk copy of large tool outputs described in #636, and the one with quadratic write growth.

The file had two runtime consumers, and both only read the condensed form (tool name, reasoning, duration, outcome byte count, artifact filename):

• continuation-prompt rendering (load_tool_traces_for_plan → artifact/tool-chain lines the coordinator sees)
• the run manifest's TaskSummary.tool_trace (session history, last_tool_call error context)

So instead of deleting the data those features need, PersistenceWrapper::on_complete now records a ToolTraceEntry into an in-memory per-task map on ExecutionPersistence, and both consumers read from that. Behavior of continuation prompts and manifests is unchanged; the on-disk file, the full-output duplication, and ToolCallRecord are gone. Full outputs remain available through promoted artifacts and OTel spans.

One intentional loss: tool call arguments were stored in ToolCallRecord.arguments but never read by anything; they are now only visible in OTel traces.

Prompt journal

The AURA_PROMPT_JOURNAL-gated prompt-journal.md recorded every coordinator/worker prompt. OTel content recording (OTEL_RECORD_CONTENT) captures the same prompts on the agent.stream/agent.turn spans, and persistence still writes per-worker prompt.txt files. Removed the module, the env flag, and the orchestrator's current_iteration atomic (its only reader was the journal).

Testing

• cargo test -p aura -p aura-config -p aura-events -p aura-cli: all green (1078 aura lib tests, incl. rewritten trace tests)
• cargo +nightly fmt --check, cargo clippy --workspace, make lint-commits via the runner image: clean
• Also fixes the pre-existing web-server lib test compile break on nightly (the final_item test helper was missing the new cache_usage field); cargo test -p aura-web-server --lib now passes (167 tests).

Neither surface is documented on the hosted docs (both were dev diagnostics), so no mezmo/documentation change should be needed; worth a grep there for AURA_PROMPT_JOURNAL before merge if unsure.

[greptile-apps]

Greptile Summary

This PR removes redundant prompt-journal and per-attempt tool-call persistence while retaining condensed tool traces in memory for continuation prompts and run manifests.

• Replaces tool-calls.json reads and writes with shared in-memory ToolTraceEntry storage.
• Removes the prompt journal, its environment flag, and iteration-tracking state.
• Updates persistence-wrapper tests to validate in-memory trace recording and fixes the web-server test initializer.

Confidence Score: 5/5

The PR appears safe to merge.

No blocking failure remains.

Important Files Changed

Filename	Overview
crates/aura/src/orchestration/persistence.rs	Replaces persisted full tool-call records with a clone-shared, in-memory condensed trace map; the previously flagged field comment is now value-focused.
crates/aura/src/orchestration/persistence_wrapper.rs	Records condensed success or error outcomes directly into execution persistence while retaining artifact promotion.
crates/aura/src/orchestration/orchestrator.rs	Removes prompt-journal integration and switches continuation and manifest construction to in-memory tool traces.
crates/aura/src/orchestration/prompt_journal.rs	Deletes the obsolete prompt-journal implementation and its tests.
crates/aura-web-server/src/streaming/handlers.rs	Updates the test-only FinalResponseInfo initializer with the required cache_usage field.

_{Reviews (2): Last reviewed commit: "test(web-server): add cache_usage to the..." | Re-trigger Greptile}