The harness was getting a 404 on /get_feature_flag, failing all 16
feature-flag tests. Implements the endpoint (mirrors posthog-ios): set
person/group properties, reload flags, resolve the value, flush.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

…n init

Declare capabilities ["capture_v0", "encoding_gzip"] in /health so the harness
runs the capture suites (android posts events to /batch and gzips them).
Also set remoteConfig = false in /init so tests don't depend on the remote
config load.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Use setGroupPropertiesForFlags(type, props, reloadFeatureFlags = false) for
group_properties (no extra reload), matching the iOS adapter pattern. group()
is still used for the type->key registration since the /flags `groups` field
reads from the GROUPS preference, which only group() writes — its
$groupidentify event and reload-on-new-group remain (no suppression param).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

…r-get-feature-flag

# Conflicts:
#	pnpm-workspace.yaml

SDK.reset() reloads flags as anon (PostHog.kt:1519 — intended behavior for
real-app logout), but that /flags lands in the next test's mock window and
inflates flag counts ("Expected 0, got 1" on no_flags_request_on_init_alone
and friends). close() has no network I/O; the next /init builds a fresh
instance. Matches the iOS adapter's pattern.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Per review: the remoteConfig=false in /init claimed to stop a /flags reload,
but PostHogRemoteConfig.kt:243 only chains to /flags when preloadFeatureFlags
is true — so the config does nothing observable here. Dropping it.

Also trims the capabilities and group() comments to one line each.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

…d.sleep

flush() is fire-and-forget with no SDK completion callback, so /flush and
/get_feature_flag slept a fixed FLUSH_SETTLE_MS to let the batch reach the
mock. Replace the blind sleep with a Condition signalled by the tracking
interceptor once the /batch response is in hand (the mock has already
recorded the request by then): returns the instant the batch lands, capped
at FLUSH_SETTLE_MS so a never-sent flush can't hang.

Addresses Manoel's review: deterministic wait, no sleeps. Safer than the
fixed sleep since we only return after the HTTP round-trip is observed,
never earlier.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

…a /flags

identify() is the only public setter for the distinctId carried in /flags,
but it unconditionally fires its own reload, adding a second /flags request
per call. That regressed 7 harness tests that assert exact /flags counts
(flags_request_*, groups_round_trip, two_flag_calls_produce_two_remote_requests,
mock_response_value_is_returned_to_caller, get_feature_flag_captures_*).

Revert to the deterministic single explicit reload. distinct_id is left
decoded-but-unapplied (same as disable_geoip/force_remote): it cannot be set
without forcing a second reload, so request_with_person_properties_device_id
stays a known public-API gap rather than breaking 7 other tests.

Verified locally against posthog-sdk-test-harness: 38/45, matching baseline
with no regression.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>