docs: fix cross-doc inconsistencies (tool count, env vars, anchors, EN/ES parity)

[scops]

Summary

Fixes documentation inconsistencies found in a cross-doc audit (README, docs/**, .github, examples). Verified against the real code before each change. Docs-only — no production code changes (one MCP server module docstring updated). Both EN and ES doc copies kept in sync.

Tool count → 14

Multiple docs claimed 11/12/13 MCP tools; the server actually exposes 14. Corrected in:

• engrama/adapters/mcp/server.py module docstring ("eleven" → "fourteen", added engrama_status, engrama_reindex, engrama_gdpr_forget)
• docs/architecture.md / .es.md ("Thirteen" → "Fourteen", added the missing engrama_gdpr_forget entry)
• docs/backends.md / .es.md, docs/vision.md / .es.md, docs/roadmap.md / .es.md, examples/claude_desktop/system-prompt.md

Stale "not yet on PyPI" caveats

vision, backends and index.es said Engrama wasn't on PyPI and to install from source — contradicting the README (PyPI badge + pip install engrama quick start). Updated to pip install engrama. Left DDR-004's design-time note as a historical record.

Env vars documented but unused in code

• HYBRID_ALPHA / HYBRID_GRAPH_BETA — never read; superseded by the ENGRAMA_FUSION_MODE / ENGRAMA_RRF_K / ENGRAMA_GRAPH_* family. Replaced in .env.example and the architecture config tables with the real knobs.
• ENGRAMA_PROFILE — never read; the schema profile is the --profile CLI flag. Removed from .env.example and the architecture / index config tables.
• NEO4J_USER → NEO4J_USERNAME (code uses NEO4J_USERNAME) in DDR-003 EN/ES example blocks.
• README security note: EMBEDDING_PROVIDER=null → none (canonical value).

EN/ES parity

docs/index.md (EN) was missing the Codex and ChatGPT Desktop MCP-integration subsections that index.es.md already had. Ported them and restructured the heading to ## MCP integration with ### Claude Desktop / ### Codex / ### ChatGPT Desktop.

Broken anchors

• index.md: internal link updated for the renamed heading (#mcp-integration-claude-desktop → #mcp-integration).
• backends.es.md: #capa-de-protocolo-y-backends → #capa-de-protocolos-y-backends (heading is plural).
• index.es.md: two accented anchors (#configuración-..., #referencia-de-configuración) → ASCII-folded forms that mkdocs actually generates.

Minor

• .github/ISSUE_TEMPLATE/bug.yml: repro placeholder used a non-existent engrama remember CLI command and engrama init without the required --profile; replaced with valid commands.

Notes / left as-is (intentional)

• DDR-003/DDR-004 design-time references to HYBRID_* and "not yet on PyPI" are point-in-time decision records — left intact (except the NEO4J_USER typo).
• docs/changelog.md "10 tools" entry is a historical release note — left intact.
• docs/glosario.md is an acknowledged EN translation stub — out of scope.

Test plan

• grep sweep confirms no residual HYBRID_ALPHA/HYBRID_GRAPH_BETA/ENGRAMA_PROFILE/NEO4J_USER= in current-facing docs
• No residual non-14 tool counts (except the intentional historical roadmap/changelog notes)
• All edited anchors verified against actual headings in the target files