feat: add haskell to kad-dht interop matrix

[adust09]

Add haskell to the Kad-DHT suite.

libp2p-hs is a Haskell implementation of
the libp2p stack. It is already registered in the transport suite as
haskell-v0.1 (#134); this adds its Kademlia DHT node here.

Test app

The node implements all three roles behind the ROLE env var and follows the
contract in kad-dht/README.md:

• Reads ROLE, TEST_KEY, REDIS_ADDR
• bootstrap — listens, SETs {TEST_KEY}_bootstrap_addr, runs a DHT server
• provider — dials the bootstrap, runs provide + putValue, then SETs
  {TEST_KEY}_provider_done
• querier — waits for both keys, runs findProviders + getValue, prints
  status: pass / status: fail (plus error: lines) on stdout
• Keys match the py and dotnet nodes: interop-test-key-{TEST_KEY} and
  /example/data/{TEST_KEY}, value hello from haskell client
• All logging goes to stderr; stdout carries only the YAML result

The image builds from the repository's own Dockerfile with the repo root as
build context, so it uses the existing source.dockerfile path in
build-images.sh — the same mode as nim. No changes to build-images.sh,
generate-tests.sh, or run-single-test.sh were needed.

Validation

Full local run of all 37 matrix tests involving haskell, with the image built
from the pinned commit through the same path CI takes:

→ Building kad-dht-haskell from adust09/libp2p-hs@25cf16e3 (-f interop/kad-dht-node/Dockerfile)
Ran: 37, Skipped: 27, Passed: 37, Failed: 0

Getting there took two fixes in libp2p-hs (adust09/libp2p-hs#265), both found by
running against this suite rather than against itself:

	passed
initial	8 / 37
+ store/match the hello from payload convention	30 / 37
+ register Identify handlers	37 / 37

The second one is worth flagging for anyone adding an implementation:
dotnet-libp2p awaits /ipfs/id/1.0.0 on every outbound connection before its
dial resolves, so a node that does not serve Identify is undialable from dotnet
even though the transport handshake completes.

Note on the existing matrix

For reference, the 27 pre-existing py/dotnet/nim tests score 24/27 on the
same machine. The three failures (py_x_py_x_dotnet, dotnet_x_py_x_dotnet,
dotnet_x_dotnet_x_dotnet) are dotnet-querier lookups timing out and are
unrelated to this change — they reproduce without haskell in images.yaml.

The matrix grows from 27 to 64 tests. The image is a GHC build and takes roughly
10 minutes from cold; let me know if that is a problem for the test pass budget.