irept deserialisation: avoid unnecessary construct/destruct

[tautschnig]

Do not allocate empty irept elements to incrementally replace them, but instead
construct them bottom-up. When running on SV-COMP's ReachSafety-ECA, these
changes avoid 805,690,478 calls to irept::detach and 554,309,582 calls to
irept::remove_ref in read_irep. In reference_convert, the number of
irept::remove_ref calls is approximately halved (from 1,209,156,897 down to
637,698,242), and the number of irept copy constructor calls reduces from
302,984,665 to 56,124.

• Each commit message has a non-empty body, explaining why the change was made.
• n/a Methods or procedures I have added are documented, following the guidelines provided in CODING_STANDARD.md.
• n/a The feature or user visible behaviour I have added or modified has been documented in the User Guide in doc/cprover-manual/
• Regression or unit tests are included, or existing tests cover the modified code (in this case I have detailed which ones those are in the commit message).
• My commit message includes data points confirming performance improvements (if claimed).
• My PR is restricted to a single feature or bugfix.
• n/a White-space or formatting changes outside the feature-related changed lines are in commits of their own.

[allredj]

✔️
Passed Diffblue compatibility checks (cbmc commit: ba4a005).
Build URL: https://travis-ci.com/diffblue/test-gen/builds/100569562

[smowton]

Numbers of calls aside, what is the actual speedup?

[tautschnig]

Numbers of calls aside, what is the actual speedup?

The number of overall invocations within the set timeout increases from 4230 to 4237, and the amount of time spent deserialising decreases by about 20 seconds (~20%), but all this is below measurement accuracy. It's mainly an effort to eradicate any unnecessary remove_ref or detach calls to remove noise from the profile.