gateway: take the farm channel role from the slot, not the farm name (ibx#253)

[userFRM]

Summary

• connect_farm derived the farm's channel role from a literal US name (src/gateway.rs:510): if farm_id == "ushmds" { "2" } else { "1" }. Every regional HMDS name fell to the else and requested channel 1, the trading channel, for a historical-data transport.
• The file already supports those names — parse_farm_route handles regional routes with a test for zdc1.ibllc.com/euhmds/4000, and the comments at :1197 and :1607 state that EU accounts route to eufarm/euhmds/secdefeu. The role check was the one place that still assumed US.
• The choice is now farm_channel_id, keyed on the hmds suffix, so ushmds, euhmds and aphmds all take channel 2 and everything else stays on 1.
• Closes gateway: farm channel role is derived from the literal name "ushmds", so a regional HMDS (euhmds) is established on the trading channel #253.

Test

hmds_farms_take_the_data_channel_in_every_region pins the mapping for all three HMDS regions plus usfarm, eufarm and secdefeu. It is a free function beside parse_farm_route, so the mapping is testable without standing up a socket — which is why it had no coverage before.

cargo test --lib — 803 pass, clippy clean. The two config::expiry_tests failures are present on main before this branch.

Not verified live

I have a US-routed account, so the EU path is unexercised here. The change is a pure widening: ushmds keeps the exact value it had, and only names that previously got the trading channel by accident are affected.

Test plan

• cargo test --offline --lib — 803 passed. The 2 failures are config::expiry_tests::{named_zone_converts_with_dst, instant_round_trips_to_wire}, which fail on the base commit too: the host has no legacy timezone files (fixed separately in config: resolve the legacy timezone names IB states its times in (ibx#335) #336).
• cargo check --offline --features python — clean.

🤖 Generated with Claude Code

[userFRM]

Reworked. The first revision keyed the role off the farm name, and the name is not a reliable discriminator.

cashhmds is the counterexample, and it is already in this repo (tests/ib_paper_compat/connection.rs:47): it ends in hmds but the extra-farm path connects it on the trading slot. My suffix rule would have moved it to channel 2, so the fix traded one wrong classification for another. Routing tags also carry whatever the server sends, so no naming invariant exists to lean on.

The caller never needed to guess. connect_farm already receives the slot as its last argument — 18 from the trading call site, 17 from the market-data one (gateway.rs:1652, :1656, and the two reconnect paths). Reading that is correct for every farm name, including both euhmds and cashhmds. The two literals are now named constants.

Also fixed: my previous insertion split parse_farm_route's doc comment, leaving the parser undocumented and its doc describing the new helper. That was a plain mistake, now restored.

Still not established by this repo: that channel 2 is right for regional HMDS specifically. Tag 6171 assigns slot 17 for every region, so it follows that the routing request should match — but the only direct evidence for the value 2 is the original ushmds branch. If you have a capture of a regional account's farm logon, that settles it.