Adopt artifact-first context for judgment-independent agents

[fuseraft]

Summary

Extends the Context: spec (artifact-first assembly) to the remaining judgment-independent
roles across the fuseraft init templates, adds an empty_sources/context_strategy signal
so a declared source that resolves to nothing is visible instead of silently empty, and removes
two pieces of dead weight: the long-deprecated EnableMemory agent field (a no-op since memory
became always runtime-injected) and the experimental, unused repl-next command.

Type of change

• Bug fix
• New feature
• Plugin
• Config example
• Documentation
• Refactor / cleanup
• Other:

Related issues

No related issue.

Changes

• Applied Context: specs to the Reviewer/Auditor/Verifier-equivalent agents in the audit,
  brownfield, devops, devteam, and research init templates (previously ContextWindow
  filtering or full shared history), so these roles assemble from durable artifacts rather than
  replayed chat.
• ContextAssembler.AssembleForAgentAsync now returns an AgentContextAssembly that reports
  which declared sources resolved to no content (EmptySources), instead of just the message list.
• AgentOrchestrator, GraphOrchestrator, and MagenticOrchestrator all emit the new
  context_strategy, declared_sources, and empty_sources fields on the context_assembly
  event; the context window visualization surfaces empty_sources as a warning line.
• Removed the obsolete EnableMemory field from AgentConfig (superseded by KnowledgeWeight;
  had no effect once ContextAssemblyPipeline went always-on), its backward-compat merge logic
  in OrchestratorBuilder, and all mentions in docs/configuration.md, docs/context-management.md,
  and config/examples/fuseraft-designer.yaml.
• Removed the experimental repl-next command (ReplNextCommand, ReplNextTurn) and its
  FUSERAFT_REPL_NEXT default-entrypoint switch from Program.cs — it duplicated ReplCommand's
  setup/turn-loop behind a hidden command with no callers or docs depending on it.
• Dropped a couple of stray "cost" mentions (README.md's VS Code sessions panel description,
  AgentMessage.Usage doc comment) that referred to an estimate not actually surfaced there.

Testing

• Added or updated unit tests
• All tests pass locally (./build.sh --target=Test)
• Tested manually end-to-end
• Documentation updated where relevant
• config/examples/ updated if config schema changed

Manual verification: fuseraft init --template audit --no-interactive generates agent files
with working Context: blocks and no EnableMemory remnants; fuseraft --help no longer lists
repl-next, and fuseraft repl --help is unaffected.

Invariants

• Execution order unchanged (Selection → Validation → Failure handling → Termination → Iteration cap)
• Any new file/shell/git tool is wrapped by ChangeTracker
• Any new validator is deterministic, side-effect free, and idempotent
• Physical history is not stripped or reordered outside of compaction

Notes for reviewers

EnableMemory and repl-next removal are pure deletions of dead code — neither had any effect
or caller left, so there's no behavior change to review there beyond "does it still compile and
pass tests." The substantive review surface is the Context: spec adoption in the init templates
and the new empty_sources signal on the context_assembly event.