feat(framework): repo files as persistent AI memory (#260)

[suleimansh]

Closes #260. Part of #204 (Open Loop). Rom: systematically use special repo files as persistence + AI memory.

We already write DECISIONS.md. This adds the read side: the agent reads the repo memory files at the start of a run for context, and is told to keep the ones it owns current.

• loadRepoMemory(cwd) reads the canonical files present (CODE-OVERVIEW.md, KNOWLEDGE-BASE.md, BRAINSTORMING.md, DECISIONS.md).
• runFramework gains a memory option; the files are framed into the system prompt alongside personas + skills: current contents (context) + a maintain instruction (persistence). The agent has file access, so it does the writing.
• CLI reads the files from the workspace and passes them in, same pattern as signals + the-framework.yml.

Correctness: DECISIONS.md is framework-owned (we write it from the decisions ledger at run end), so it is framed read-only. Otherwise the agent could edit it and our end-of-run write would clobber that.

Verified e2e with --fake --cwd <dir with memory files>:

◆ project memory: CODE-OVERVIEW.md, KNOWLEDGE-BASE.md
# no files -> no line

New exports: loadRepoMemory / memoryFraming / MEMORY_FILES / MemoryFile / LoadedMemory. Typecheck + 113 framework tests green. Changeset minor.

Follow-ups (separate): a summarize/update step on the ledger, and truncation for large files.