Connectors: fetch command type and payload in a single round-trip (getCommandInfo)

[biohazardxxx]

Summary

Note: this branch is stacked on #308 (payload truncation fix) and includes its commits — the new work is the last 6 commits. Merge #308 first and this PR reduces to its own changes.

Fetching one command currently costs two blocking cross-thread round-trips into the connector's io thread: getCommandType (posts a packaged_task, blocks) then getPayload inside the handler (posts another, blocks). This PR merges them into one:

• MtService: new GetCommandInfo(handle, int& type, std::string& payload) — a single packaged_task that pops the next command AND returns its payload. Existing methods untouched; everything touching expert state still runs on the single io thread.
• Both connectors export:

  // pops the next command; returns required payload length in wchar_t (excl. null) or -1 on error;
  // copies at most capacity-1 chars (always null-terminated); type receives the command type
  int getCommandInfo(int expertHandle, int& type, wchar_t* payload, int capacity, wchar_t* err);

• Both EAs (MQL5 + MQL4) fetch type and payload in one call into a module-level _command_payload; GetJsonPayload() parses that instead of calling the DLL — all Execute_* handlers untouched. Oversized payloads re-fetch via getPayload2 (safe: GetCommandPayload reads current_task_ non-destructively; the command is not re-popped).
• getCommandType/getPayload/getPayload2 remain exported and functional — old .ex4/.ex5 binaries keep working with the new DLLs (live-verified, see Testing).

Implementation notes

• Adversarial review found one medium defect: a payload conversion failure after the pop (e.g. invalid UTF-8) silently dropped the command, leaving the client waiting the full CommandTimeout where the old path returned an immediate error. Fixed in 364b193: the EA now dispatches with an empty payload so the handler emits the same immediate "Failed to get payload" error response as before.
• Review also confirmed: single packaged_task (no nested post/deadlock), empty-queue loop exits identical to the old path, stdcall decoration correct on Win32 (_getCommandInfo@20), no stale-payload leakage across commands.

Testing

• Both connectors build clean (x64 + Win32); getCommandInfo present in both DLLs' export tables (dumpbin). Both EAs compile Result: 0 errors, 0 warnings; recompiled .ex5/.ex4 included.
• Runtime-verified against a live MT5 terminal (build 5830) with the new DLL + new EA:
  • Short and 6016-char Print (exercising the oversized re-fetch path): both return true.
  • 50 sequential commands: 0.5–0.6 ms/command end-to-end through the merged path.
  • Backward compat: two experts running the previous .ex5 on the same new DLL answered commands normally via the old exports.
• MQL4 side is source-identical and compiles clean but was not runtime-tested against an MT4 terminal.

🤖 Generated with Claude Code