Refactor: Removes Rc from Refcell<AccountSharedData> in the program-runtime

[Lichtso]

Problem

Continuation of #21882.

• Feature: TransactionContext, InstructionContext and BorrowedAccount #21706 will require the keyed_accounts parameter of mock_process_instruction() to be split in transaction_accounts and instruction_accounts.
• The tuple of the keyed_account parameter should be replaced by the struct AccountMeta and instruction_accounts (see discussion in #20448).
• Individual test cases should be isolated and the accounts generated by one and consumed by another explicitly routes through an accounts variable.
• Tests currently can mock impossible transactions, which have two different accounts for the same key or two different keys for the same account. The transaction_accounts mapping should always be bijective.

Summary of Changes

• Removes Rc from Refcell<AccountSharedData> in the program-runtime
• Makes TransactionError::AccountBorrowOutstanding obsolete, but does not remove it yet
• Adjust all program tests accordingly

Fixes #

[codecov]

Codecov Report

Merging #21927 (6d19ba1) into master (46e5350) will increase coverage by 0.0%.
The diff coverage is 98.9%.

@@           Coverage Diff            @@
##           master   #21927    +/-   ##
========================================
  Coverage    81.2%    81.3%            
========================================
  Files         516      516            
  Lines      144353   144777   +424     
========================================
+ Hits       117315   117772   +457     
+ Misses      27038    27005    -33     

[jackcmay]

Do we need pubkey in both of these vectors? Would an index suffice?

[Lichtso]

An index would suffice for the instruction_accounts, but then AccountMeta would have to go.
So, as that is only for the tests / mocks I think we can live with that inefficiency for now.
Once #21706 has landed I can extend its structures to the tests as well.

[jackcmay]

Is this same paradign going to be extended beyond mock? If so, are we going to be allocating a lot of vectors as we pass this information around?

[Lichtso]

No, the Vec allocation only happens this way for the tests / mocks, and only for now, as it might be refined to become index based as well (see comment above). The rest of the program-runtime will be directly migrated to the new structs defined in #21706 and they work based on references and indices, so there should not be more allocation happening than what we had until now.