Fix large PPLNS Stratum job handling

[Roalkege]

Summary

Fixes an issue where NerdMiner remained at 0 KH/s when connected to pools that send large Stratum mining.notify
messages, such as public-pool.io:13333.

These PPLNS jobs can contain a significantly larger coinbase transaction than typical solo-pool jobs.

Root cause

Two size limitations prevented these jobs from being processed correctly:

1. The fixed 4 KB ArduinoJson document was too small to classify and parse large mining.notify messages.
2. Coinbase data was copied into a fixed 512-byte buffer, truncating coinbase transactions longer than 511 characters.

As a result, large jobs were either discarded during JSON parsing or hashed using incomplete coinbase data.

Changes

• Classify Stratum messages using an ArduinoJson filter so large unused fields are not copied into the JSON document.
• Parse mining.notify messages in zero-copy mode.
• Store Merkle branches independently of the temporary JSON input buffer.
• Hash the complete coinbase transaction incrementally in bounded chunks.
• Avoid unnecessary copies of large mining-job structures.
• Release large coinbase strings after the block header has been prepared.
• Fix an out-of-bounds null terminator in the Merkle-root buffer.

Testing

Tested with:

• Pool: public-pool.io
• Port: 13333
• Device: LilyGO T-Display-S3
• PlatformIO environment: NerdminerV2

Results:

• Large PPLNS jobs are parsed successfully.
• Multiple subsequent mining.notify jobs are accepted.
• No watchdog resets or crash loops.
• Stable hashrate of approximately 247–253 KH/s.
• Currently the PPLNS Pool ignores the suggest_difficulty command and we get diff 1 Jobs :/

The changes also compile successfully for:

• NerdminerV2 — ESP32-S3
• ESP32-2432S028R — classic ESP32

Before

Parsing of the large mining.notify message failed
0 KHashes
0.00 KH/s

## After

Parsing Method [MINING NOTIFY]
merkle sha: ...
Completed 12020 Khashes, avg. hashrate 245.5 KH/s