feat(dashboard): improve java-tron Grafana dashboards usability

[warku123]

What does this PR do?

Improves the java-tron Grafana dashboard templates with more useful panels, expression fixes, and removal of panels that don't work well in practice.

java-tron-server.json (46 → 69 panels):

• Adds 24 new panels across existing rows: Tx TPS breakdown (broadcast/block), Block fork count, Block miss change, SR block latency, Block process latency from broadcast, Block fetch latency, Transaction process latency by smart contract, a low-transaction-block table (blocks with ≤20 txs per miner in the selected range), per-DB level panels, JVM memory usage rate, storage overview, etc.
• Fixes Transaction Process Latency panel: switches irate → rate (irate only uses the last two scrape points and produces spiky, noisy curves), restores the type="block" filter (without it, every contract renders two same-named series — the trx broadcast pre-execution path and the block in-block path — making the legend unreadable), and adds group/instance scoping.
• Fixes Block Push Latency panel: the average query was missing its / rate(..._count[...]) divisor, so the "avg" series was actually avg × block rate.
• Reworks SR Set panels: expands the SR Set row and changes both panels from cumulative-since-process-start to $__range-windowed semantics (increase(...)[$__range]), so they show SR changes within the selected time range instead of monotonically growing counters; drops the arbitrary 5/10 thresholds.
• Removes panels that don't hold up in practice: P95/P99 Block Transaction Count (per-miner quantiles degenerate to single-sample noise — each SR produces ~1 block per 81s, shorter than the typical rate window — and the 300/500/2000 bucket granularity makes P99 interpolation meaningless), Avg Transactions per Block (redundant with Transaction Cnt Per Block), 4 legacy Transaction TPS panels (replaced by the TPS breakdown), the old Block Process Latency panel (replaced by "From Broadcast"), and the System row (OS metrics are covered by the Node Exporter Full dashboard).
• DB size panels: drops the type="LEVELDB" filter (9 panels) so RocksDB-backed instances render data.
• TCP Traffic Rate/s: adds group/instance scoping to avoid cross-instance legend mixing.

java-tron-api.json / java-tron-mechanism.json: instance variable now cascades from group (label_values({group="$group"}, instance)); api panel 99 scrape interval fixed from 1s to 60s.

java-tron-api-statistic.json: adds an instance variable for parity with the other dashboards.

Why are these changes required?

Several existing panels have expression bugs (missing divisor, missing filters, noisy irate) or rely on metric semantics that don't produce useful output (per-miner quantiles, cumulative counters). The new panels cover gaps operators frequently need: per-miner block quality, TPS composition, DB growth per level, and SR set churn.

This PR has been tested by:

• Live Prometheus queries against mainnet fullnode/SR instances confirming every new/fixed panel returns expected data
• JSON validity + layout checks (no duplicate panel ids, no overlapping grid positions)

Extra details

• All PromQL queries follow the existing dashboard conventions ($group, $instance, $__rate_interval).
• All panels use metric families already exported by java-tron (tron:block_transaction_count, tron:sr_set_change, tron:db_size_bytes, etc.) — no java-tron changes required.