faces/totp: improve memory usage

The TOTP face is working in the simulator but fails on the real hardware
when loaded with lots of codes, just like the LFS version.
This is likely caused by the recent refactoring of the TOTP face
which introduced a declarative credential interface for ease of use.
That's accomplished by decoding the secrets at runtime which increases
the RAM requirements. Users are likely hitting memory limits.

In order to mitigate this, the algorithm is changed from decoding
all of the secrets only once during initialization to on the fly
decoding of the secret for the current TOTP credential only.
This converts this face's dynamic memory usage from O(N) to O(1)
at the cost of memory management when switching faces and credentials
which could impact power consumption. Issue is confirmed fixed by
author of issue who has tested it on real hardware. Fixes #384.

Due to variable key sizes, the memory cannot be statically allocated.
Perhaps there's a maximum key size that can serve as worst case?

Also took this opportunity to restructure the code a bit.
Also added code to check for memory allocation failure.

Reported-by: madhogs <59648482+madhogs@users.noreply.github.com>
Fixed-by: Matheus Afonso Martins Moreira <matheus.a.m.moreira@gmail.com>
Tested-by: Matheus Afonso Martins Moreira <matheus.a.m.moreira@gmail.com>
Tested-on-hardware-by: madhogs <59648482+madhogs@users.noreply.github.com>
Signed-off-by: Matheus Afonso Martins Moreira <matheus.a.m.moreira@gmail.com>
GitHub-Issue: https://github.com/joeycastillo/Sensor-Watch/issues/384

movement/watch_faces/complication/totp_face.h

@@ -70,6 +70,8 @@ typedef struct {
     uint8_t steps;
     uint32_t current_code;
     uint8_t current_index;
+    uint8_t *current_decoded_key;
+    size_t current_decoded_key_length;
 } totp_state_t;
 
 void totp_face_setup(movement_settings_t *settings, uint8_t watch_face_index, void ** context_ptr);