Save descriptor template to NVRAM during signing on Nano S; general data structure refactoring

[bigspider]

Some wallet policies that were below the size limit as a string were parsed to a memory representation larger than the maximum memory limit. This is undesirable, as the second limit is implementation specific and hard to deal with for developers.

As increasing the limit without further optimization is not an option on Nano S, because almost all the available RAM is used during the signing flow, this approach adds an optimization (only on Nano S, and only during the signing flow): move the wallet policy to the NVRAM during the signing flow, freeing up several hundreds of bytes of RAM. This allows to increase the limits while gaining some extra slack in the RAM, which will be very useful for the new features.

For most common wallet policies, the descriptor template is smaller than 64 bytes, resulting in a single nvram page write for each signing flow. The maximum size of 320 results in at most 5 nvram page writes for very complex policies. As a single installation of the bitcoin app amounts to ~1000 page writes, that can be considered negligible.

The difficulty for this change was that the abstract syntax tree that is built while parsing the descriptor template contains pointers; therefore, relocating it is not trivial.
The solution is to use relative pointers (an offset w.r.t. to their own position) instead of absolute pointers. This was already done for some pointers in order to save memory (using 2 bytes instead of 4 for such pointers); this PR extends it to all the pointers in the structure, making the AST free to relocate in memory (and possibly saving a few more bytes). The approach to relative pointers was also heavily refactored, generalizing it and improving the type safety of relative pointers; moreover, bound checks and null-pointer checks were added, which makes them a lot safer to use and the code more maintainable.

Closes: #134

[codecov-commenter]

Codecov Report

All modified and coverable lines are covered by tests ✅

Comparison is base (d6b3f56) 84.88% compared to head (0aec12c) 84.85%.

Additional details and impacted files

@@             Coverage Diff             @@
##           develop     #216      +/-   ##
===========================================
- Coverage    84.88%   84.85%   -0.04%     
===========================================
  Files           17       17              
  Lines         2210     2192      -18     
===========================================
- Hits          1876     1860      -16     
+ Misses         334      332       -2     

Flag	Coverage Δ
unittests	84.85% <100.00%> (-0.04%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[sonarcloud]

Quality Gate failed

Failed conditions

63.2% Coverage on New Code (required ≥ 80%)
9.6% Duplication on New Code (required ≤ 3%)

See analysis details on SonarCloud