fix: arweave mempool and raw request errors handling

[charmful0x]

findings

• to_tx_message/6 could crash with badmatch on non-not_found data fetch errors due to {ok, Data} = ...
• invalid /raw ids could throw cannot_encode_path (err 500)
• non-numeric pending chunk offsets could crash with badarg in hb_util:int/1 (err 500)

patches

• to_tx_message/6 now do (as it was doing before impr: mempool Arweave transaction headers and data support #842 ):
  - {ok, RawData} -> tx with data
  - {error, not_found} -> header only tx
  - other errors -> returned as Error

• high level guard head_raw/3 with ?IS_ID, so invalid /raw ids return not_found instead of throwing cannot_encode_path (vs https://arweave.net/raw/hello)

• wrap pending offset parsing in pending/3, so non-numeric offsets return not_found instead of crashing with badarg

tests

• tx_raw_fetch_error_round_trips_test
• `head_raw_invalid_id_returns_not_found_test
• pending_invalid_offset_returns_not_found_test

and all tests pass

 [done in 85.576 s]
=======================================================
All 38 tests passed.

note

i noticed that pending offsets still have one edge case:

• non-numeric offsets like offset=abc now resolve locally as 404 (as described in the patch)
• BUT, negative offsets like offset=-1 still parse and round-trip upstream arweave nodes, which currently returns {"error":"invalid_offset"}

^

should both be canonalized? hyperbeam mimicking arweave ideally in both cases?

also should i target feat/mempool-offsets instead of neo/edge @samcamwilliams ?

followup

• pending/3 offset validation now return error shape equal to arweave node and 400 status code

http://tip-1.arweave.xyz:1984/unconfirmed_chunk/aa/-1

[charmful0x]

closing this and moving to work cleanly on #857 targeting feat/mempool-offsets