Implement the TODOs in the Harvard Architecture PR (#711)

In the [Harvard Architecture
PR](#688) we left a few TODOs.
Here we make good on them.

There's a few things happening in this PR:

- We remove the fake instruction `EANY`. `EANY` was only ever an
artifact of Risc0's incredible 'interesting' approach to decoding. We
use the real instructions `ECALL` and `EBREAK` instead.

- Remove `AUIPC` and `LUI`. Both of them are now implemented as
pseudo-instructions, that translate to `ADDI` during decoding.

- Use the same library as SP1 for RiscV decoding, instead of
copy-and-pasting-and-editing Risc0's 'interesting' decoder. That
simplifies our code, and comes with a lot more tests than we ever had.
Both because of explicit tests in the library, and because of the usage
in SP1 and other projects. This gets rid of much error prone bit
manipulating code.

- Use `struct Instruction` throughout the code when handling and testing
instructions, instead of `u32`. That makes specifying tests a lot
simpler and more readable. No more
`0b_000000001010_00000_000_00001_0010011, // addi x1, x0, 10` in the
code.

- Remove the notion of executable vs non-executable ROM. This is only
necessary for a von-Neumann architecture: everything that's in our
instruction-cache is meant to be executable already. (We can
re-implement this restriction later by controlling what is allowed to
make it into the instruction cache when we eg decode the ELF. But it's
unnecessary: we already honour the executable flag for memory sections
in the ELF.)

ceno_emul/Cargo.toml

@@ -15,6 +15,7 @@ elf = "0.7"
 itertools.workspace = true
 num-derive.workspace = true
 num-traits.workspace = true
+rrs_lib = { package = "rrs-succinct", version = "0.1.0" }
 strum.workspace = true
 strum_macros.workspace = true
 tracing.workspace = true