refactor: move distributed awareness to PGBL

[Becheler]

This is a cross-repo change and comes with a companion PR in PBGL : boostorg/graph_parallel#47

In short: When Boost was modularized, the PBGL part moved to the graph_parallel repository, but the forwarding hooks, use_mpi.hpp, and the MPI build wiring remained in graph. This PR closes the migration story.

Before submitting

• This PR targets the develop branch.
• I searched for an existing PR or issue covering the same change.
• My contribution is licensed under the Boost Software License 1.0.

Type of change

• Bug fix
• New feature or API addition
• Refactor (no behavior change)
• Documentation
• Build, CI, or tooling
• Other (specify below)

Does this PR introduce a breaking change?

• Yes (describe migration impact below)
• No

What this PR does

• Deleted use_mpi.hpp, detail/mpi_include.hpp, detail/empty_header.hpp
• Stripped the forwarding hooks from all 11 algorithm headers:
  • top mpi_include.hpp include
  • trailing BOOST_GRAPH_MPI_INCLUDE(...)),
  • the BFS distributed/concepts.hpp forward
  • the #ifdef BOOST_GRAPH_USE_MPI bfs_helper declaration block.
• Removed the three MPI regions from pending/property_serialize.hpp (kept the two generic serialize() overloads).
• Removed import mpi from build/Jamfile.v2

Motivation

• PBGL was historically developed inside libs/graph. To let one code path run sequentially or distributed, the sequential headers included their distributed counterparts when BOOST_GRAPH_USE_MPI. was defined. When Boost was modularized, PBGL moved to the graph_parallel repository, but the forwarding hooks, use_mpi.hpp, and the MPI build wiring remained in graph.
• So the dependency was unfortunately inverted: BGL names and its build depends on its own downstream consumer, pulling MPI and Boost.Serialization into its dependency graph.
• In 2018, commit 5f1edca5 (Peter Dimov) wrapped the includes in a BOOST_GRAPH_MPI_INCLUDE macro to hide them from boostdep: the reported dependency disappeared but the actual coupling did not (build/Jamfile.v2 still runs import mpi).
• The forwarding feature itself was actually partially broken today with Fruchterman–Reingold: distributed/fruchterman_reingold.hpp uses point_traits but no header on any include path pulls in boost/graph/point_traits.hpp. So #define BOOST_GRAPH_USE_MPI + #include <boost/graph/fruchterman_reingold.hpp> has never compiled in any release. No distributed test covers this so it went unnoticed.

The user contract (remains unchanged with this PR):

"include <boost/graph/use_mpi.hpp> first, then the serial algorithm header, get the distributed version for free":

It allows to write code like:

#include <boost/graph/use_mpi.hpp> // first, always
#include <boost/graph/distributed/mpi_process_group.hpp>
#include <boost/graph/distributed/adjacency_list.hpp>
#include <boost/graph/breadth_first_search.hpp> // sequential header only

using namespace boost;

typedef adjacency_list<vecS,
    distributedS<graph::distributed::mpi_process_group, vecS>,
    undirectedS> Graph;

int main(int argc, char* argv[])
{
    mpi::environment env(argc, argv);
    Graph g(100);
    // runs the distributed BFS, its header never included by name
    breadth_first_search(g, vertex(0, g), visitor(default_bfs_visitor()));
}

• The file keeps the same path <boost/graph/use_mpi.hpp> but it ships from graph_parallel instead of graph. Existing includes resolve unchanged.
• the forwarding mechanism is preserved: the relocated use_mpi.hppincludes all the distributed algorithm headers, a user who includes only the sequential <boost/graph/breadth_first_search.hpp> and calls it on a distributed graph still gets the distributed overload.
• Users who already include distributed/… headers explicitly are of course unaffected.

Testing

This was locally verified: 5 in-tree PBGL tests/examples that rely on this pattern compiled with zero edits.

Checklist

• Existing tests pass (b2 in the test/ directory).
• New behavior is covered by a test, or this is a docs / build / refactor change.
• Documentation was updated if user-facing behavior changed.
• No new compiler warnings on the platforms I built against.

[github-actions]

Boost dependency footprint vs develop (auto-generated).
PR run 33504971594 vs develop run 33494674376 (417011fa53).

Header-inclusion weights (graph files pulling each direct dependency in):

No header-inclusion-weight changes.

Transitive Boost modules: 66 → 66 (0)

[github-actions]

Compiler-warning counts vs develop (auto-generated).
PR run 33504971560 vs develop run 33494674338 (417011fa53).

Job	Baseline	After	Delta
macos (clang, 14)	648	648	0
macos (clang, 17)	611	611	0
macos (clang, 20)	611	611	0
ubuntu (clang-19, 14)	648	648	0
ubuntu (clang-19, 17)	611	611	0
ubuntu (clang-19, 20)	611	611	0
ubuntu (clang-19, 23)	611	611	0
ubuntu (gcc-14, 14)	801	801	0
ubuntu (gcc-14, 17)	937	937	0
ubuntu (gcc-14, 20)	937	937	0
ubuntu (gcc-14, 23)	937	937	0
windows_msvc_14_3 (msvc-14.3)	961	961	0

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.