rpcv2 integration tests: run the rpcv1 suite unchanged against rpcv2, four CI legs

[karthikiyer56]

Part of #986.

Goal

Run the rpcv1 integration suite, assertions unchanged, against rpcv2. CI runs four legs instead of two.

leg	daemon	protocol
1	rpcv1	P27
2	rpcv1	P28
3	rpcv2	P27
4	rpcv2	P28

• The only thing that may differ between the two daemons is how the harness instantiates them.
• rpcv1's integration-test settings do not change.
• rpcv2 starts with the stock defaults from cmd/stellar-rpc/rpcv2/rpc-v2-sample-config.toml.
• Run model stays as today: one Core container per test, captive core as a subprocess, the daemon in-process.

This replaces the earlier in-process paritytest design. rpcv2 now serves every rpcv1 method natively (#908, #960), so the suite itself is the parity test.

Layout

path	holds
cmd/stellar-rpc/internal/integrationtest/	the common tests and infrastructure/ (moved up from rpcv1/)
cmd/stellar-rpc/internal/rpcv1/integrationtest/	rpcv1-only tests
cmd/stellar-rpc/internal/rpcv2/integrationtest/	home for rpcv2-only tests (rpcv2 backfill, getEventsV2), empty except doc.go

• The harness picks the daemon from STELLAR_RPC_INTEGRATION_TESTS_DAEMON (rpcv1 default, rpcv2).
• Each CI leg runs the common package and its own daemon's package.

Which tests run where

test	placement	reason
TestGetLedgersFromDatastore, backfill_test.go, migrate_test.go, ingest_loadtest_test.go	rpcv1-only	mutate rpcv1/config.Config, read SQLite, or run released rpcv1 images
TestArchiveUserAgent	rpcv1-only	pins rpcv1's User-Agent; rpcv2 has none yet (follow-up below)
TestHealth	common, minus one line	LedgerRetentionWindow == 17280 moves to a per-daemon test; rpcv2 measures retention in 10 000-ledger chunks
TestUpgradeFrom20To21	common	probe changes from GetDaemon().GetDB() to getLatestLedger; assertions unchanged
TestMetrics	common	after rpcv2 registers the same process metrics (below)
everything else	common	unchanged

rpcv2 code changes (test-enabling only)

• Export a minimal options entry point wrapping runDaemonWith: logger and flag overrides. The binary's behaviour does not change.
• Register soroban_rpc_build_info, the logmetrics log-line counters, and the Go and process collectors on rpcv2's registry, as rpcv1/daemon/metrics.go does.

rpcv2 instantiation in the harness

• Config: a test-only TOML derived from the sample, [backfill.datastore] removed, admin endpoint on, preflight debug on. Per-test values (data dir, endpoint, ports, captive-core file, archive URL, core binary) go in as flag overrides. A unit test parses it strictly.
• Ports: RPC, admin, captive core admin HTTP, captive core query, captive core peer, all from getFreeTCPPorts, with a retry on bind collision.
• Captive core file: the harness appends NETWORK_PASSPHRASE for rpcv2 only. The shared template stays byte-identical.
• sendTransaction goes through captive core's admin port, rpcv2's stock route. First implementation step is a spike of TestSendTransactionSucceeds under rpcv2.
• Every wait is sized for four environments booting at once, as in test: run every integration environment in the parallel batch #981.
• TestGetFeeStats is the first end-to-end check of rpcv2's fee windows against rpcv1's. A mismatch is an rpcv2 bug to fix, not an assertion to change.

CI

• integration-tests.yml gains a daemon input and runs the common package plus the daemon's package.
• stellar-rpc.yml replaces the two hand-written jobs with one caller job and strategy.matrix over protocol × daemon. Job names: Integration tests (P27, rpcv1) and so on. The protocol list moves with releases.
• continue-on-error stays as it is for all four legs.