docs: add Memory concept page for agents (#29890)

* docs: add Memory and Skills concept pages for agents

- memory.mdx: covers four types of memory (read-only context, writable
  short-form, searchable context, on-demand skills), how agents interact
  with memory via auto-generated tools, system prompt structure, and
  gotchas around prompt caching and compaction overlays.

- skills.mdx: covers on-demand document loading via SkillProvider,
  R2SkillProvider usage, uploading/managing skills, generated tools
  (load_context, unload_context, set_context), custom providers, and
  comparison with other memory types.

* docs: combine skills into memory page as a dedicated section

* docs: rewrite memory types to be use-case driven

- Read-only: traditional system prompt / SOUL.md, loaded from code or R2
- Writable: scratchpad like Claude Code's todo list, agent-maintained notes
- Searchable: large knowledge bases, agent retrieves chunks via search
- Loadable (Skills): whole documents loaded on demand, not small chunks

* docs: clarify searchable context is provider-agnostic (Vectorize, FTS, etc.)

* docs: remove non-Cloudflare provider mentions

* docs: add [writable] tag to writable block examples

* docs: collapse Skills section into Loadable context (Skills), remove duplication

* docs: add conversation history as primary memory type, restructure as context memory, add micro-compaction

* docs: clarify session storage is backed by a Session Provider, defaulting to SQLite

* docs: replace em dashes with commas/colons/periods throughout

* docs: use WranglerConfig component for R2 bucket binding

Author: mattzcarey