port speller perf fix from 4.8

[SamBent]

Addresses Issue #3350. This is a port of a servicing fix in .NET 4.7-4.8.

Description

PR #2871 added logic to the word-breaking phase to extend misspelled tokens ("words") with non-token characters ("punctuation") if such an extension fixes the misspelling. I've found ways to speed it up:

1. The logic checks each token and potential extension for misspellings, using an existing method ComprehensiveCheck that automatically populates each spelling error with a list of suggestions obtained from the native layer. These suggestions are never used, and the native calls were accounting for 80% of the time (says Trevor Fellman). The logic only cares whether any errors exist, so use a new method HasErrors that answers that question without asking for suggestions.

2. The check for potential extensions is skipped if the non-token characters that follow the token are all white-space. In practice, those characters often include nulls ('\0') at the end. Removing these nulls first allows us to skip the extension test altogether in many cases - about 40% in my experiments.

3. Don't consider trailing whitespace or nulls in the non-token characters. (This generalizes the original heuristic that discarded the non-token characters if they were all whitespace.)

4. Don't consider whitespace or nulls interior to the non-token characters, until reaching a character that is not whitespace or null.

5. Cache the 10 most recent HasError results, and answer queries from the cache instead of calling the native spell-checker.

This speeds up the spell-checker quite a bit, but it is still slower than the original. The word-breaking phase still has to check each token for misspelling (which it didn't do before #2871); there's no getting around this expense.

Customer Impact

This bug is blocking migration to .NET Core.

Regression

Regression in .NET 5.0.

Testing

Ad-hoc with customer scenarios.
Standard regression testing.

Risk

Low. Straightforward port of .NETFx fix that was released last year.