Nanox so fix

[ImplOfAnImpl]

The reason why tests in feature/refactor-too-big crash is stack overflow, which happens because some large objects are passed/returned/moved by value. I fixed the crash by boxing them.

I documented my findings about the apps stack usage in docs/memory_usage.md (in particular, how to measure it), but in short: passing, returning or even moving around large objects eats up large amounts of stack space and currently we only have less than 12KB of it on nanox.

Details:
sizeofs of some objects in feature/refactor-too-big:
TxParsingInputsContext: 688
SignTxReq: 264
mlcp::SighashInputCommitment: 208
TxInputReq: 264
TxOutputReq: 192
Response: 100
NbglStreamingReview: 48

I boxed those with sizes around 200 bytes or more.

Original stack frame sizes (sorted by frame size, bigger first):

7792    mintlayer_app::handlers::sign_tx::handle_sign_tx
3952    mintlayer_app::handle_command
1384    sample_main
1336    mintlayer_app::app_ui::utils::bech32m_encode
880     <mintlayer_core_primitives::tx_output::TxOutput as parity_scale_codec::codec::Decode>::decode::<&[u8]>
592     ledger_device_sdk::exiting_panic
512     mintlayer_app::app_ui::sign::format_lock
384     <alloc::collections::btree::map::entry::Entry<mintlayer_app::handlers::sign_tx::summary_collector::CoinOrTokenId, mintlayer_core_primitives::misc::Amount>>::or_insert
360     mintlayer_app::handlers::utils::mintlayer_hash
328     <ledger_device_sdk::io_legacy::Comm>::decode_event::<ledger_device_sdk::io_legacy::ApduHeader>
304     <ledger_device_sdk::io_legacy::Comm>::apdu_send
288     <ledger_device_sdk::hash::blake2::Blake2b_512 as ledger_device_sdk::hash::HashInit>::new
144     <ledger_device_sdk::ecc::Secp256k1 as ledger_device_sdk::ecc::SeedDerive>::derive_from
136     <mintlayer_core_primitives::crypto::PublicKey as parity_scale_codec::codec::Decode>::decode::<&[u8]>
128     <mintlayer_core_primitives::misc::Amount as parity_scale_codec::codec::Decode>::decode::<&[u8]>
112     mintlayer_app::handlers::sign_message::schnorr_sign::<32>
104     compiler_builtins::mem::memmove
104     <ledger_device_sdk::nbgl::nbgl_streaming_review::NbglStreamingReview>::next
104     <u128>::_fmt
104     mintlayer_app::app_ui::sign::format_value
...

A sidenote: one interesting thing is that if I marked fn decode_all as #[inline(never)], then I could see:

1440	messages::decode_all::<messages::SignTxReq>

I.e. more than 1 KB of stack usage above may be attributed to the inlined decode_all (not that we can do much about it, just an observation).

After boxing each variant of TxParsingContext I got:

3152    mintlayer_app::handlers::sign_tx::handle_sign_tx
2384    mintlayer_app::handle_command
1384    sample_main
...

Then I boxed by-value self in XxxContext methods:

2520    mintlayer_app::handlers::sign_tx::handle_sign_tx
2384    mintlayer_app::handle_command
1384    sample_main

Then I boxed the contents of SignTxReq and made the correspondent handle_ functions accept Box:

1896	mintlayer_app::handlers::sign_tx::handle_sign_tx
1648	mintlayer_app::handle_command
1384	sample_main
...

After removing "chain-error" from parity-scale-codec's features, the compiler decided to inline handle_command and I got:

2984	sample_main
1896	mintlayer_app::handlers::sign_tx::handle_sign_tx
1336	mintlayer_app::app_ui::utils::bech32m_encode
...

But note that the new total size of sample_main and handle_sign_tx stack frames is less than the old total size of sample_main+handle_command+handle_sign_tx.

For clarity I also recompiled it with handle_command marked with #[inline(never)]:

1896	mintlayer_app::handlers::sign_tx::handle_sign_tx
1544	mintlayer_app::handle_command
1384	sample_main
...

So removing "chain-error" reduced handle_command's stack usage by about 100 bytes. Not a big deal, but on the other hand we don't benefit from "chain-error" in any way, because we never examine decoding errors and don't even print them.

For comparison, frame sizes in feature/mintlayer-app are:

4792	mintlayer_app::handle_command
2112	sample_main
1336	mintlayer_app::app_ui::utils::bech32m_encode
1008	<mintlayer_core_primitives::tx_output::TxOutput as parity_scale_codec::codec::Decode>::decode::<&[u8]>
...

P.S.

1. The tests in this branch still fail when run without --golden_run, because the snapshots are outdated.
2. I also bumped ledger_secure_sdk_sys version to 1.16.1 (same as in feature/mintlayer-app) - this is just to make results in the two branches comparable.