build: handle architecture include more gracefully

The "openssl/opensslconf.h" header is copied from an
architecture-specific source file, which means that specific path is
not valid within the source tree. I previously hacked around this by
copying the file within the source tree, but that had the major
downside of invalidating various cache layers after the copy was
performed.

Since we install this header with the right name, a slightly better
solution, hopefully, is to add the header install path as an include
directory. In theory, this will not invalidate any caching and improve
the build process slightly.

build.zig

@@ -243,13 +243,6 @@ pub fn libresslBuild(
         else => @panic("unsupported target CPU arch"),
     };
 
-    try b.build_root.handle.copyFile(
-        conf_header,
-        b.build_root.handle,
-        source_header_prefix ++ "openssl/opensslconf.h",
-        .{},
-    );
-
     libressl_libs.libcrypto.installHeader(conf_header, "openssl/opensslconf.h");
     libressl_libs.libssl.installHeader(conf_header, "openssl/opensslconf.h");
     libressl_libs.libtls.installHeader(conf_header, "openssl/opensslconf.h");
@@ -296,6 +289,13 @@ pub fn libresslBuild(
         else => @panic("unsupported target CPU arch"),
     }
 
+    // add the header install path to the include path so that compilation will pick
+    // up "openssl/opensslconf.h". This is added last to avoid interfering with the
+    // somewhat messy include handling that libressl does.
+    libressl_libs.libcrypto.addIncludePath(.{ .path = b.getInstallPath(.header, "") });
+    libressl_libs.libssl.addIncludePath(.{ .path = b.getInstallPath(.header, "") });
+    libressl_libs.libtls.addIncludePath(.{ .path = b.getInstallPath(.header, "") });
+
     libressl_libs.libssl.linkLibrary(libressl_libs.libcrypto);
 
     // cmake builds libtls with libcrypto and libssl symbols jammed into it. However,