chore: remove dir_upload_public / dir_download_public surface (daemon + 15 SDKs + docs)

[Nic-dorman]

Summary

`dir_upload_public` / `dir_download_public` is a daemon-fabricated surface with no upstream backing. Upstream `WithAutonomi/ant-client` v2 has no directory primitive in `ant-core/src/data/client/` — only `file_upload` / `file_download` and variants (`file_upload_with_mode`, `file_prepare_upload`, `file_prepare_upload_with_visibility`, `finalize_upload`, `file_download_with_progress`, `estimate_upload_cost`). The daemon's dir handler called `file_upload_with_mode` on a directory path, which fails with `HTTP 500 — encryption failed: I/O error: Is a directory (os error 21)` because the underlying primitive expects a file. The dir surface was never functional in v2; it was only latent because previous example stubs errored at `file_cost` first.

This PR removes the surface end-to-end. Mirror of the graph_entry removal precedent. If a consumer later needs native directory upload, file it as a real product feature with a defined chunking + manifest contract.

What changes (18 commits)

1. Daemon (`830d6e4`)

• `antd/src/rest/files.rs` — drop `dir_upload_public` and `dir_download_public` handlers
• `antd/src/rest/mod.rs` — drop `/v1/dirs/upload/public` and `/v1/dirs/download/public` route registrations
• `antd/src/grpc/service.rs` — drop `dir_upload_public` + `dir_download_public` gRPC methods
• `antd/proto/antd/v1/files.proto` — drop `DirUploadPublic` + `DirDownloadPublic` RPCs (shared message types stay)
• `antd/src/types.rs` — drop `DirUploadPublicResponse` struct
• `antd/openapi.yaml` — drop the two dir paths + `DirUploadPublicResponse` schema
• `antd/README.md` — drop endpoint-table rows

2. SDK fan-out (15 per-SDK commits) — `antd-go`, `antd-py`, `antd-js`, `antd-rust`, `antd-cpp`, `antd-csharp`, `antd-java`, `antd-dart`, `antd-ruby`, `antd-php`, `antd-elixir`, `antd-lua`, `antd-zig`, `antd-kotlin`, `antd-swift`. Each drops:

• The sync and async client methods (`dir_upload_public`, `dir_download_public`, async variants where present)
• The gRPC client wrappers
• Any dir-specific result models
• The exercising unit tests (or test cases within larger files)
• README rows describing the methods

3. MCP wrapper (`5b0c230`) — `antd-mcp/src/antd_mcp/server.py`: drop `is_directory` flag from `upload_file` / `download_file` MCP tools.

4. Docs (`9d2b513`) — quickstart-*.md for cpp/csharp/dart/elixir/java/kotlin/lua/php/python/ruby/rust/swift/zig + `llms.txt` + `llms-full.txt` cleaned of dir-upload examples and method-signature rows.

What does NOT change

• Generated proto stubs (`_pb.rb`, `.pb.go`, `*_pb2_grpc.py`, etc.) — they'll clear on next proto regen. Same deferral pattern as the graph removal precedent. Nothing imports the dead generated types.
• Shared proto messages (`UploadFileRequest`, `DownloadPublicRequest`, `DownloadResponse`) — used by file routes too, stay.
• File upload/download surface — completely untouched. Consumers wanting to upload a directory can recurse over files client-side and use `file_upload_public` per file.
• Examples — the canonical `04-files` examples were already file-only (separate prior fix); this PR doesn't re-touch them.

Why now

v1.0 ship target. Shipping v1.0 with a daemon route that 500s on the happy path + 15 SDKs that ship a method which 500s is bad. Removal is the right call until a real product spec emerges. `-285 / +0` on the daemon alone; full PR ~`-1500 lines` of dead code across the repo.

Test plan

• `git grep -E "dir_upload|dir_download|dirUpload|dirDownload|DirUpload|DirDownload"` returns only generated proto stubs (acceptable, deferred) + one FFI README reference (out of scope, separate concern — FFI is a planned UniFFI binding to the `autonomi` Rust crate, not the daemon).

• Per-SDK unit tests green where toolchain available locally: go, python, js, rust, csharp, java, dart, php (all 100% pass).

• Daemon compiles clean (`cargo check`).

• End-to-end dev2 sweep on a 5-PR integration branch (this PR + the four in-flight example/cleanup PRs):

  SDK	Result	SDK	Result
  python	✅ 13s	lua	✅ 11s
  go	✅ 11s	zig	✅ 51s
  js	✅ 13s	dart	✅ 19s
  rust	✅ 19s	csharp	✅ 27s
  cpp	✅ 50s	swift	✅ 159s
  ruby	✅ 12s	java	✅ 78s
  php	✅ 11s	kotlin	✅ 98s
  elixir	✅ 13s

  15/15 PASS against `ant dev start --enable-evm` running the canonical 5 examples (connect/data/chunks/files/private) per SDK.

• Reviewer: re-run the sweep on a fresh devnet, or re-merge to confirm the merge state matches.

Out of scope

• Implementing native directory upload as a real feature — separate product spec needed.
• Proto-stub regen across SDKs — deferred per the graph_entry removal precedent. The generated dead types don't break anything; they'll clear on next regen.
• The one FFI README reference to dir_upload — flagged in the subagent report; out of scope here since FFI is a different planned-API doc, not a daemon caller.

🤖 Generated with Claude Code