fix: update foxy for Boost 1.90 compatibility and unpin macOS Boost

[kinyoklion]

Summary

Updates the vendored foxy library to compile against Boost 1.90 (which removed three boost::beast::http::field enum values) while retaining backward compatibility with Boost 1.8x. Unpins macOS Homebrew Boost from the deprecated boost@1.85 so it installs the latest version.

Foxy change: Adds a #if BOOST_VERSION >= 109000 guard in export_connect_fields.hpp around the hop_by_hops array. Boost 1.90 removed proxy_authentication_info, proxy_features, and proxy_instruction from boost::beast::http::field, so the Boost 1.90+ path uses an 8-element array without those fields. The pre-1.90 path is unchanged (11 elements).

Boost unpin: Reverts brew install boost@1.85 → brew install boost in install-boost/action.yml, so macOS CI builds against the latest Homebrew Boost (currently 1.90). This removes the boost@1.85 pin introduced in PR #505, which had a Homebrew deprecation deadline of 2026-04-05.

Both #if branches are validated by CI: macOS builds use Boost 1.90 (#if path), Linux uses 1.81 and Windows uses 1.87 (#else path). All 36 CI checks pass.

Review & Testing Checklist for Human

• Unpinning strategy: Unpinning means macOS CI will track Homebrew's latest Boost. Future Boost updates could introduce new breakages in vendored foxy. Confirm this is acceptable vs. pinning to boost@1.90.

Test plan: CI validates both preprocessor branches across all three platforms. The macOS jobs passing confirm Boost 1.90 compatibility (including that proxy_connection was not removed alongside the other three fields). Linux and Windows jobs confirm the #else path still compiles correctly with older Boost versions.

Notes

• Requested by: @kinyoklion
• Link to Devin Session

---

Note

Medium Risk
CI will now track Homebrew’s latest Boost on macOS, which can introduce future build breakages. The code change is a small compile-time guard but affects HTTP header filtering behavior across Boost versions.

Overview
macOS CI now installs Homebrew’s default boost package instead of the pinned boost@1.85, so builds track the latest Boost release.

The vendored foxy header export_connect_fields.hpp is updated for Boost 1.90+ by conditionally excluding the removed boost::beast::http::field enum values (proxy_authentication_info, proxy_features, proxy_instruction) via a BOOST_VERSION >= 109000 guard, preserving the previous hop-by-hop header list for older Boost versions.

^{Written by Cursor Bugbot for commit 5442186. This will update automatically on new commits. Configure here.}

[devin-ai-integration]

🤖 Devin AI Engineer

I'll be helping with this pull request! Here's what you should know:

✅ I will automatically:

• Address comments on this PR. Add '(aside)' to your comment to have me ignore it.
• Look at CI failures and help fix them

Note: I can only respond to comments from users who have write access to this repository.

⚙️ Control Options:

• Disable automatic comment and CI monitoring

[kinyoklion]

Devin, update with the latest changes from main.

[devin-ai-integration]

Done — merged latest main (which includes the merged PR #505) into this branch. Waiting for CI now.

[kinyoklion]

Devin check your Review & Testing Checklist for Human against CI and remove items covered by CI.

[devin-ai-integration]

Done — reviewed all checklist items against CI and removed the ones that are covered:

• proxy_connection validity — removed. macOS CI builds with Boost 1.90 and the 8-element array compiles successfully; if proxy_connection were removed, the build would fail.
• Behavioral differences — removed. Contract tests pass on all platforms, covering HTTP header handling behavior.
• No other foxy files reference removed fields — removed. macOS builds with Boost 1.90 would fail to compile if any other file referenced the removed fields; all pass.

Only remaining item is the unpinning strategy question (whether to track latest Boost vs. pin to 1.90), which is a policy decision CI can't validate.

[kinyoklion]

aside: The brew installs of boost from homebrew are rapidly deprecated. So when we pin we have to constantly update them, and we also don't get a preview of what will soon be breaking issues.