ffi: fix float passing in libffi for riscv64

[kxxt]

libffi was incorrectly passing floats as doubles on RISC-V. This PR applies the merged upstream fix (libffi/libffi#972) for the vendored copy of libffi since libffi appears to have a very long release cycle (last release was at 2025-08-02).

This fixes the following three new test failures caught in https://github.com/riscv-forks/node-riscv/actions/runs/27544719447/job/81437248188:

out/Release/node --experimental-ffi --expose-gc --test-reporter=./test/common/test-error-reporter.js --test-reporter-destination=stdout /home/kxxt/node-riscv/test/ffi/test-ffi-dynamic-library.js
out/Release/node --experimental-ffi --expose-internals --test-reporter=./test/common/test-error-reporter.js --test-reporter-destination=stdout /home/kxxt/node-riscv/test/ffi/test-ffi-shared-buffer.js
out/Release/node --experimental-ffi --expose-gc --test-reporter=./test/common/test-error-reporter.js --test-reporter-destination=stdout /home/kxxt/node-riscv/test/ffi/test-ffi-calls.js

CC @nodejs/platform-riscv64

I am testing this PR on riscv64 at https://github.com/riscv-forks/node-riscv/actions/runs/27756747452/job/82120430409?pr=4

[nodejs-github-bot]

Review requested:

• @nodejs/ffi
• @nodejs/security-wg

[kxxt]

It appears that there is some problem with NixOS binary cache that causes CI to fail.

error: hash mismatch importing path '/nix/store/bl7rmhhsy7vjb9qm3jfwgqpv3cn7wfb1-openssl-3.6.2';
         specified: sha256:0nagwij5x8yi54380yw53xy1ksn9jclp98lxmizbqq6hgngqp553
         got:       sha256:1fypgxjwksn6kh2ig4fmaskfnflicjq60zf5il2s9nxblhkkvnww
error: path '/nix/store/bl7rmhhsy7vjb9qm3jfwgqpv3cn7wfb1-openssl-3.6.2' is required, but there is no substituter that can build it

[mcollina]

lgtm

[kxxt]

I have rebased this PR to fix the flaky test in the CI (#63923).

[ShogunPanda]

I think this will be handled by #64040 which seems to include this fix. Do you think we can close this?

[kxxt]

I think this will be handled by #64040 which seems to include this fix. Do you think we can close this?

Yes. It’s great to see a new libffi release just a few days after my fix was merged, considering the last release happened 10 month ago. I am closing this PR in favor of #64040.